Kev-4B: Qwen3.5-4B-Base, decision-v7 recipe (locked test 0.870 / 0.832); previous Qwen3 weights at tag qwen3
Browse files- README.md +45 -35
- adapter_config.json +6 -1
- adapter_model.safetensors +2 -2
- head.pt +1 -1
- provenance.json +20 -18
- result.json +0 -0
- tokenizer.json +2 -2
- tokenizer_config.json +22 -229
- train.log +318 -316
- training_config.json +5 -5
- training_metrics.json +5 -5
README.md
CHANGED
|
@@ -2,7 +2,7 @@
|
|
| 2 |
language: en
|
| 3 |
license: apache-2.0
|
| 4 |
library_name: peft
|
| 5 |
-
base_model: Qwen/Qwen3-4B-Base
|
| 6 |
base_model_relation: adapter
|
| 7 |
pipeline_tag: text-classification
|
| 8 |
tags:
|
|
@@ -11,7 +11,7 @@ tags:
|
|
| 11 |
- lora
|
| 12 |
- multiple-choice
|
| 13 |
- typesafe
|
| 14 |
-
-
|
| 15 |
datasets:
|
| 16 |
- legacy-datasets/banking77
|
| 17 |
- google/boolq
|
|
@@ -31,73 +31,83 @@ model-index:
|
|
| 31 |
- name: Kev-4B
|
| 32 |
results:
|
| 33 |
- task: { type: text-classification, name: typed decision (choice / noul / score) }
|
| 34 |
-
dataset: { type: mixed, name: "decision-
|
| 35 |
metrics:
|
| 36 |
-
- { type: accuracy, value: 0.
|
| 37 |
-
- { type: expected_calibration_error, value: 0.
|
| 38 |
- task: { type: text-classification, name: typed decision, out-of-domain }
|
| 39 |
dataset: { type: mixed, name: "transfer-v4 development (764 records; six never-trained sources + held-out policy structures)" }
|
| 40 |
metrics:
|
| 41 |
-
- { type: accuracy, value: 0.
|
| 42 |
-
- { type: brier_score, value: 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
---
|
| 44 |
|
| 45 |
# Kev-4B
|
| 46 |
|
| 47 |
-
Kev-4B is a **decision model**: one document (the *state*) and a set of typed questions in, a probability distribution per question out, in one forward pass. No text generation. It is a LoRA adapter (r=16) plus a pointer head on `Qwen/Qwen3-4B-Base`, serving TypeSafe's public `/v1/systemone` contract.
|
| 48 |
|
| 49 |
-
**The recommended
|
| 50 |
|
| 51 |
-
- Hub: `jaredpalmer/kev-4b` (this repo; trial `
|
| 52 |
-
- Code, suites, every trial with hashes and paired bootstraps: [github.com/jaredpalmer/kev](https://github.com/jaredpalmer/kev) — `PLAN.md`, `runs/leaderboard.md`
|
| 53 |
|
| 54 |
## Results (same frozen items for every row)
|
| 55 |
|
| 56 |
-
| | Kev-
|
| 57 |
-
|---|---|---|---|---|
|
| 58 |
-
| in-distribution accuracy (decision-
|
| 59 |
-
| out-of-domain accuracy (transfer-v4 dev,
|
| 60 |
-
| out-of-domain Brier | 0.
|
| 61 |
-
| confident errors out of domain (p ≥ 0.9 and wrong) |
|
| 62 |
-
|
|
| 63 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 64 |
|
| 65 |
-
|
| 66 |
|
| 67 |
-
|
| 68 |
|
| 69 |
-
**
|
| 70 |
|
| 71 |
-
## What
|
| 72 |
|
| 73 |
-
- **
|
| 74 |
-
- **
|
| 75 |
-
-
|
| 76 |
-
- Programmatic contrastive policy pairs teach the trained rule structures (both-correct 0.85–1.0) but transfer to unseen structures only partially (0.5–0.6 at 4B, 0.03–0.11 at 0.6B).
|
| 77 |
|
| 78 |
## Known limits
|
| 79 |
|
| 80 |
-
-
|
| 81 |
-
-
|
| 82 |
-
-
|
| 83 |
-
-
|
|
|
|
| 84 |
|
| 85 |
## Training
|
| 86 |
|
| 87 |
-
Frozen suite `evals/
|
| 88 |
|
| 89 |
## Evaluation protocol
|
| 90 |
|
| 91 |
-
Development partitions select models; the locked test partition is read at most once per candidate. Every number carries suite hash, code hashes
|
| 92 |
|
| 93 |
## Use
|
| 94 |
|
| 95 |
```bash
|
| 96 |
-
uv run --extra serve python -m kev.serve --run jaredpalmer/kev-4b --port 8008 # KEV_DTYPE=bf16 on a
|
| 97 |
```
|
| 98 |
|
| 99 |
Any TypeSafe-compatible client works: `TypeSafeClient(api_key="local", base_url="http://127.0.0.1:8008", model="kev-latest")`.
|
| 100 |
|
| 101 |
## License
|
| 102 |
|
| 103 |
-
Apache-2.0 for the adapter and head; Qwen3 base is Apache-2.0; datasets carry their own licenses.
|
|
|
|
| 2 |
language: en
|
| 3 |
license: apache-2.0
|
| 4 |
library_name: peft
|
| 5 |
+
base_model: Qwen/Qwen3.5-4B-Base
|
| 6 |
base_model_relation: adapter
|
| 7 |
pipeline_tag: text-classification
|
| 8 |
tags:
|
|
|
|
| 11 |
- lora
|
| 12 |
- multiple-choice
|
| 13 |
- typesafe
|
| 14 |
+
- qwen3.5
|
| 15 |
datasets:
|
| 16 |
- legacy-datasets/banking77
|
| 17 |
- google/boolq
|
|
|
|
| 31 |
- name: Kev-4B
|
| 32 |
results:
|
| 33 |
- task: { type: text-classification, name: typed decision (choice / noul / score) }
|
| 34 |
+
dataset: { type: mixed, name: "decision-v7 development (1,204 records; ten trained public sources + programmatic policy data)" }
|
| 35 |
metrics:
|
| 36 |
+
- { type: accuracy, value: 0.877 }
|
| 37 |
+
- { type: expected_calibration_error, value: 0.059, name: "ECE, raw probabilities" }
|
| 38 |
- task: { type: text-classification, name: typed decision, out-of-domain }
|
| 39 |
dataset: { type: mixed, name: "transfer-v4 development (764 records; six never-trained sources + held-out policy structures)" }
|
| 40 |
metrics:
|
| 41 |
+
- { type: accuracy, value: 0.794 }
|
| 42 |
+
- { type: brier_score, value: 0.316 }
|
| 43 |
+
- task: { type: text-classification, name: typed decision, out-of-domain, locked test }
|
| 44 |
+
dataset: { type: mixed, name: "transfer-v4 test (read once)" }
|
| 45 |
+
metrics:
|
| 46 |
+
- { type: accuracy, value: 0.832 }
|
| 47 |
+
- { type: brier_score, value: 0.266 }
|
| 48 |
---
|
| 49 |
|
| 50 |
# Kev-4B
|
| 51 |
|
| 52 |
+
Kev-4B is a **decision model**: one document (the *state*) and a set of typed questions in, a probability distribution per question out, in one forward pass. No text generation. It is a LoRA adapter (r=16, 33.8M trainable parameters) plus a pointer head on `Qwen/Qwen3.5-4B-Base` (revision `1001bb4d`), serving TypeSafe's public `/v1/systemone` contract.
|
| 53 |
|
| 54 |
+
**The recommended Kev.** The best accuracy per byte: out of domain 0.794 on the development partition and **0.832 on the locked test**, against the Qwen3 Kev-4B's 0.790 / 0.806 on the same items, with better calibration (Brier 0.266 vs 0.294 on the test) and the highest held-out-rule score of any 4B (0.78). Same recipe at four seeds: transfer 0.788 / 0.800 / **0.794** / 0.770, held-out pairs 0.72 / 0.75 / 0.78 / 0.69; this checkpoint is seed 2, selected on the development partition (highest development accuracy).
|
| 55 |
|
| 56 |
+
- Hub: `jaredpalmer/kev-4b` (this repo, main revision; trial `q35-4b-s23/00-trial-0`). The previous Qwen3 checkpoint is at revision `qwen3` and has [its own card](kev-4b-qwen3.md).
|
| 57 |
+
- Code, suites, every trial with hashes and paired bootstraps: [github.com/jaredpalmer/kev](https://github.com/jaredpalmer/kev) — `PLAN_Qwen35.md` (the port and this experiment), `PLAN.md`, `runs/leaderboard.md`
|
| 58 |
|
| 59 |
## Results (same frozen items for every row)
|
| 60 |
|
| 61 |
+
| | Kev-4B (Qwen3) | Kev-8B (Qwen3) | **Kev-4B** | Kev-9B | Jev |
|
| 62 |
+
|---|---|---|---|---|---|
|
| 63 |
+
| in-distribution accuracy (decision-v7 dev, 1,204 records) | 0.854 | 0.863 | **0.877** | 0.876 | 0.845 |
|
| 64 |
+
| out-of-domain accuracy (transfer-v4 dev, 764 records) | 0.790 | 0.796 | **0.794** | 0.812 | 0.857 |
|
| 65 |
+
| out-of-domain Brier | 0.328 | 0.337 | **0.316** | 0.291 | 0.211 |
|
| 66 |
+
| confident errors out of domain (p ≥ 0.9 and wrong) | 8.2% | 9.9% | 8.2% | 7.5% | 3.7% |
|
| 67 |
+
| coverage at ≤ 5% error (share of decisions automatable) | 0.31 | 0.45 | 0.54 | 0.53 | 0.70 |
|
| 68 |
+
| held-out policy structures, both siblings correct | 0.73 | 0.69 | **0.78** | 0.80 | 0.86 |
|
| 69 |
+
| option-order flip rate | 0.06 | 0.00 | 0.08 | 0.03 | 0.00 |
|
| 70 |
+
| **locked test**, out-of-domain accuracy / Brier | 0.806 / 0.294 | 0.780 / 0.327 | **0.832 / 0.266** | 0.837 / 0.243 | – |
|
| 71 |
+
| **locked test**, in-distribution accuracy | 0.856 | 0.870 | 0.870 | 0.873 | – |
|
| 72 |
+
|
| 73 |
+
Per-source out-of-domain accuracy (Kev-4B / Jev): QNLI 0.93 / 0.93, SciQ 0.99 / 0.99, TweetEval-offensive 0.74 / 0.81, PAWS 0.74 / 0.79, MMLU 0.70 / 0.90, Emotion 0.54 / 0.59, deadline (3-level date arithmetic) 0.55 / 0.93, (A or B) and C 0.91 / 0.91, (A and B) or not C 0.88 / 0.97, if A then not B else C 1.00 / 0.78.
|
| 74 |
|
| 75 |
+
**Paired against the Qwen3 Kev-4B on the same items** (record-clustered bootstrap): development +1.3 pp [−1.3, +4.6]; **locked test +2.9 pp [−0.9, +6.4]**, Brier −0.028. The pre-registered criteria for this experiment asked for a development-partition CI excluding zero and `deadline` ≥ 0.75; neither was met (deadline 0.55). The locked read, taken once after selection, is the confirmatory number. Both facts are in `PLAN_Qwen35.md` §10.
|
| 76 |
|
| 77 |
+
**Newer evaluation columns** (`transfer-v9` development, Kev-4B / Qwen3 Kev-4B / Jev): MMLU-Pro (10-way) 0.500 / 0.440 / 0.840; state buried among unrelated records 0.66 / 0.69 / 0.70; share of *unknowable* items (deciding evidence removed) answered at ≥ 0.9 confidence 0.19 / 0.44 / 0.09 — lower is better.
|
| 78 |
|
| 79 |
+
**External suites** (same items as their published Jev numbers; measured on seed 1 of this recipe): SemIf's authored 144 — 0.896 (Qwen3 Kev-4B 0.847, SemIf's untrained Qwen3.5-4B 0.813, live Jev 0.965); scienthoon's 900 tickets — queue 0.928, angry 0.794, ECE 0.086 (Jev 0.897, 0.914, 0.105; the Qwen3 Kev-4B scored 0.687 / 0.375).
|
| 80 |
|
| 81 |
+
## What changed from the Qwen3 Kev-4B
|
| 82 |
|
| 83 |
+
- **Base model**: Qwen3.5-4B-Base, a hybrid of 24 Gated DeltaNet (linear attention) layers and 8 full-attention layers. Because the recurrent layers cannot honour a block-causal mask, questions run as separate causal rows that continue from the shared state (`kev/model.py: forward_rows_batch`); isolation is exact by construction (together vs alone within 1e-5) and on attention-only models this form is bit-identical to the packed one.
|
| 84 |
+
- **Same data and recipe** as the Qwen3 checkpoints: `decision-v7`, two epochs, LoRA r=16 (attention, MLP and DeltaNet projections), lr 5e-5. Nothing else changed, so every difference above is the base.
|
| 85 |
+
- What the new base bought: in-distribution accuracy (+2.3 pp), knowledge retention (MMLU +5 pp, MMLU-Pro +6 pp), held-out rules (0.73 → 0.78, and 3 of 4 seeds ≥ 0.70), coverage at ≤ 5% error (0.31 → 0.54), and behaviour on evidence-free questions (0.44 → 0.19 confidently answered). What it did not buy: out-of-domain accuracy beyond noise on the development partition, Emotion (0.66 → 0.54), or date arithmetic — the base does it at 0.68 zero-shot and training erodes it to 0.55 (see `PLAN_Qwen35.md`, "The deadline hypothesis").
|
|
|
|
| 86 |
|
| 87 |
## Known limits
|
| 88 |
|
| 89 |
+
- **Slow on a Mac.** The DeltaNet kernels have no MPS implementation; PyTorch falls back to reference code. A five-question request that takes 0.17 s on the Qwen3 Kev-4B takes 0.78 s here in bf16 on an M5. On CUDA with `flash-linear-attention` installed it is fast. Use `jaredpalmer/kev-4b@qwen3` for low latency on Apple Silicon until an MLX path exists.
|
| 90 |
+
- Requires `transformers >= 5.17` (the `qwen3_5` architecture) and `peft >= 0.21`.
|
| 91 |
+
- Date arithmetic (`deadline` 0.55 vs Jev 0.93), knowledge (MMLU 0.70 vs 0.90) and noisy-label emotion (0.54 vs 0.59) remain the gap to Jev. Training data that renders day counts is the next experiment ([issue #8](https://github.com/jaredpalmer/kev/issues/8)).
|
| 92 |
+
- Out-of-domain probabilities are usable but not calibrated (raw ECE 0.130 dev, 0.102 test); temperature fitted in-domain does not transfer.
|
| 93 |
+
- 4B bf16 needs ~9 GB of GPU memory for serving; training took 56 min on one H100 (peak 24.6 GB).
|
| 94 |
|
| 95 |
## Training
|
| 96 |
|
| 97 |
+
Frozen suite `evals/v7/decision-v7`: 10,000 public records (1,000 per source), 896 policy minimal-pair records over nine template families, 1,680 records from 60 randomly generated rule structures in four rendering styles. Two epochs, LoRA r=16 α=32 on `q/k/v/o_proj`, `gate/up/down_proj`, `in_proj_qkv/z/a/b`, `out_proj`; pointer head from scratch; cross-entropy on the option distribution; lr 5e-5 (OneCycle), effective batch 8, bf16 autocast with fp32 master weights, gradient checkpointing; option permutation, none-of-the-above insertion, distractors, none minimal pairs on 25% of Choice records. No Jev outputs were used for training.
|
| 98 |
|
| 99 |
## Evaluation protocol
|
| 100 |
|
| 101 |
+
Development partitions select models; the locked test partition is read at most once per candidate (`runs/locked/kev-4b-q35/`). Every number carries suite hash, code hashes and git commit in `result.json`. Untrained-base baselines use zero-shot letter logits on the same items (`scripts/base_mmlu_probe.py`).
|
| 102 |
|
| 103 |
## Use
|
| 104 |
|
| 105 |
```bash
|
| 106 |
+
uv run --extra serve python -m kev.serve --run jaredpalmer/kev-4b --port 8008 # KEV_DTYPE=bf16 on a Mac; slow on MPS, see limits
|
| 107 |
```
|
| 108 |
|
| 109 |
Any TypeSafe-compatible client works: `TypeSafeClient(api_key="local", base_url="http://127.0.0.1:8008", model="kev-latest")`.
|
| 110 |
|
| 111 |
## License
|
| 112 |
|
| 113 |
+
Apache-2.0 for the adapter and head; the Qwen3.5 base is Apache-2.0; datasets carry their own licenses.
|
adapter_config.json
CHANGED
|
@@ -3,7 +3,7 @@
|
|
| 3 |
"alpha_pattern": {},
|
| 4 |
"arrow_config": null,
|
| 5 |
"auto_mapping": null,
|
| 6 |
-
"base_model_name_or_path": "Qwen/Qwen3-4B-Base",
|
| 7 |
"bias": "none",
|
| 8 |
"corda_config": null,
|
| 9 |
"ensure_weight_tying": false,
|
|
@@ -32,11 +32,16 @@
|
|
| 32 |
"rank_pattern": {},
|
| 33 |
"revision": null,
|
| 34 |
"target_modules": [
|
|
|
|
|
|
|
| 35 |
"gate_proj",
|
| 36 |
"v_proj",
|
| 37 |
"o_proj",
|
| 38 |
"k_proj",
|
| 39 |
"up_proj",
|
|
|
|
|
|
|
|
|
|
| 40 |
"down_proj",
|
| 41 |
"q_proj"
|
| 42 |
],
|
|
|
|
| 3 |
"alpha_pattern": {},
|
| 4 |
"arrow_config": null,
|
| 5 |
"auto_mapping": null,
|
| 6 |
+
"base_model_name_or_path": "Qwen/Qwen3.5-4B-Base",
|
| 7 |
"bias": "none",
|
| 8 |
"corda_config": null,
|
| 9 |
"ensure_weight_tying": false,
|
|
|
|
| 32 |
"rank_pattern": {},
|
| 33 |
"revision": null,
|
| 34 |
"target_modules": [
|
| 35 |
+
"in_proj_b",
|
| 36 |
+
"in_proj_qkv",
|
| 37 |
"gate_proj",
|
| 38 |
"v_proj",
|
| 39 |
"o_proj",
|
| 40 |
"k_proj",
|
| 41 |
"up_proj",
|
| 42 |
+
"in_proj_z",
|
| 43 |
+
"in_proj_a",
|
| 44 |
+
"out_proj",
|
| 45 |
"down_proj",
|
| 46 |
"q_proj"
|
| 47 |
],
|
adapter_model.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:20e798d1670df3c680d9b18aa1cad85d7bd6f25d57b2eed1a1621673d04f62a9
|
| 3 |
+
size 129924032
|
head.pt
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 5248127
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0976cae95ae97515c11fc324713e29218d39556a20f6d8bcc7d0c8a16738ea27
|
| 3 |
size 5248127
|
provenance.json
CHANGED
|
@@ -1,7 +1,7 @@
|
|
| 1 |
{
|
| 2 |
"config": {
|
| 3 |
"epochs": 2,
|
| 4 |
-
"seed":
|
| 5 |
"lr": 5e-05,
|
| 6 |
"lora": 16,
|
| 7 |
"accum": 2,
|
|
@@ -18,40 +18,42 @@
|
|
| 18 |
"head_lr": 0.0,
|
| 19 |
"weight_decay": 0.01,
|
| 20 |
"anchor_w": 0.0,
|
| 21 |
-
"base": "Qwen/Qwen3-4B-Base",
|
| 22 |
"dtype": "bf16",
|
| 23 |
-
"checkpointing": 1
|
|
|
|
|
|
|
| 24 |
},
|
| 25 |
-
"config_sha256": "
|
| 26 |
"suite_sha256": "a8f50e481b7d90b97da049e0ff6a01cee2f1ed204aed61a8265af0edbb5514d2",
|
| 27 |
"source_hashes": {
|
| 28 |
"kev/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
|
| 29 |
"kev/anchors.py": "089d8a5493502bb26f540eb1c5e681780ca0bb01276073d4e1733211f15d0e10",
|
| 30 |
"kev/api.py": "cdb0602684d798ccd3fc2c86be622f2a8e7bcdeee64d07f95604c7d3fd4701ec",
|
| 31 |
"kev/autoresearch.py": "0a8aa6b57c1c9ba93d25b2cf631b03aed2e686e374c1148002eec48765b7b1fc",
|
| 32 |
-
"kev/benchmark.py": "
|
| 33 |
"kev/compare.py": "bd0445f021de59e35c7bff9304e39dd2e1211e3877a453575594ae7b81b0ada4",
|
| 34 |
"kev/composition.py": "f335ed17e18e0a544893db5e22b9a059e6ce1b2e14dbb863ac7d7bbf8f3e0536",
|
| 35 |
-
"kev/contrastive.py": "
|
| 36 |
"kev/data.py": "9729f8f497b02aca542915d6f1bea3ff55956629d7d9610370aea91ee66efd23",
|
| 37 |
-
"kev/evaluate.py": "
|
| 38 |
"kev/experiment.py": "c635394fe56c9aa11765f8bbf8f133dcf3e57cbe8f2bce4ebdc7b4e85fa536ff",
|
| 39 |
"kev/jev.py": "e0213782359ba2f95adbf045ddaf0a008b08b4162bf6a9b66d4ce55fc91a51cb",
|
| 40 |
-
"kev/model.py": "
|
| 41 |
"kev/plot.py": "d689c7dd18cde7f9ea77cb4f55c50ff7da1880b21cb2ecf244348e45842a1382",
|
| 42 |
-
"kev/publish.py": "
|
| 43 |
-
"kev/serve.py": "
|
| 44 |
-
"kev/study_v3.py": "
|
| 45 |
-
"kev/suite.py": "
|
| 46 |
-
"kev/train.py": "
|
| 47 |
-
"
|
| 48 |
-
"
|
| 49 |
-
"
|
|
|
|
| 50 |
},
|
| 51 |
-
"git_commit": "
|
| 52 |
"platform": "Linux-4.19.0-gvisor-x86_64-with-glibc2.36",
|
| 53 |
"torch": "2.8.0+cu128",
|
| 54 |
"device": "cuda",
|
| 55 |
-
"gpu": "NVIDIA
|
| 56 |
"legacy_checkpoint": false
|
| 57 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"config": {
|
| 3 |
"epochs": 2,
|
| 4 |
+
"seed": 2,
|
| 5 |
"lr": 5e-05,
|
| 6 |
"lora": 16,
|
| 7 |
"accum": 2,
|
|
|
|
| 18 |
"head_lr": 0.0,
|
| 19 |
"weight_decay": 0.01,
|
| 20 |
"anchor_w": 0.0,
|
|
|
|
| 21 |
"dtype": "bf16",
|
| 22 |
+
"checkpointing": 1,
|
| 23 |
+
"base": "Qwen/Qwen3.5-4B-Base",
|
| 24 |
+
"base_revision": "1001bb4d826a52d1f399e183466143f4da7b741b"
|
| 25 |
},
|
| 26 |
+
"config_sha256": "0e8442a3a7d49a1338a2b6600f2f11d95efb4351b21aa463e539cb3c657017c5",
|
| 27 |
"suite_sha256": "a8f50e481b7d90b97da049e0ff6a01cee2f1ed204aed61a8265af0edbb5514d2",
|
| 28 |
"source_hashes": {
|
| 29 |
"kev/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
|
| 30 |
"kev/anchors.py": "089d8a5493502bb26f540eb1c5e681780ca0bb01276073d4e1733211f15d0e10",
|
| 31 |
"kev/api.py": "cdb0602684d798ccd3fc2c86be622f2a8e7bcdeee64d07f95604c7d3fd4701ec",
|
| 32 |
"kev/autoresearch.py": "0a8aa6b57c1c9ba93d25b2cf631b03aed2e686e374c1148002eec48765b7b1fc",
|
| 33 |
+
"kev/benchmark.py": "5be816b27bc69c7244fd1b0ad619d4783b85a21b928212f306e731312f5146b2",
|
| 34 |
"kev/compare.py": "bd0445f021de59e35c7bff9304e39dd2e1211e3877a453575594ae7b81b0ada4",
|
| 35 |
"kev/composition.py": "f335ed17e18e0a544893db5e22b9a059e6ce1b2e14dbb863ac7d7bbf8f3e0536",
|
| 36 |
+
"kev/contrastive.py": "cbb979aa5d40265ad0e64695f94b281d91751fa405811ddfa8212fede111edcf",
|
| 37 |
"kev/data.py": "9729f8f497b02aca542915d6f1bea3ff55956629d7d9610370aea91ee66efd23",
|
| 38 |
+
"kev/evaluate.py": "2b15a093ae9085a2d3999d5f57cb6293ce93d101bcddd0ea252089538e8f3a07",
|
| 39 |
"kev/experiment.py": "c635394fe56c9aa11765f8bbf8f133dcf3e57cbe8f2bce4ebdc7b4e85fa536ff",
|
| 40 |
"kev/jev.py": "e0213782359ba2f95adbf045ddaf0a008b08b4162bf6a9b66d4ce55fc91a51cb",
|
| 41 |
+
"kev/model.py": "8a5db7953c4364d9d6cda29598bde2ca64e65030aabeec8087f9a160ecbc58b8",
|
| 42 |
"kev/plot.py": "d689c7dd18cde7f9ea77cb4f55c50ff7da1880b21cb2ecf244348e45842a1382",
|
| 43 |
+
"kev/publish.py": "c0b3efac3cfe93990d1846cfd306cf762a9d03e9e58380d13c2961a5ddae4619",
|
| 44 |
+
"kev/serve.py": "570e995aa98b51dbbf867694276d29fc9b9f3f1e62f22bea0116ae51d6497003",
|
| 45 |
+
"kev/study_v3.py": "9fc44d44dad09a4f1ee29a7bcb2eb3c7aa373d09186d0e9533666b69ec95c401",
|
| 46 |
+
"kev/suite.py": "f221e5a50111bab24ada035ff68ea3c3067482ff1edda48b8d4ece769043a561",
|
| 47 |
+
"kev/train.py": "e405978b6e1e7444e682a8869d1bd6c88c90bd9890836fcd84b2ab2aa0543817",
|
| 48 |
+
"kev/transfer_v9.py": "588aa2ff3ec0823c2e31733bef9a3748b263849c966ef6a80ebd686539c605e9",
|
| 49 |
+
"modal_app.py": "12c1b6c1d7f264da8644819bfd11b4baab9c541b1277fee120467514d4628591",
|
| 50 |
+
"pyproject.toml": "52da5eea3efc6f2b1c0589acebad62e56a214294bb02c1a4218c93efd4af3182",
|
| 51 |
+
"uv.lock": "18b3e5ea0f25d2e8546fab81f16cb965ae05c3289fffaaa1ce27d114adee47f3"
|
| 52 |
},
|
| 53 |
+
"git_commit": "92e6d3c9a700fbbde1b50e2d73a02fcf4157a577",
|
| 54 |
"platform": "Linux-4.19.0-gvisor-x86_64-with-glibc2.36",
|
| 55 |
"torch": "2.8.0+cu128",
|
| 56 |
"device": "cuda",
|
| 57 |
+
"gpu": "NVIDIA H100 80GB HBM3",
|
| 58 |
"legacy_checkpoint": false
|
| 59 |
}
|
result.json
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
tokenizer.json
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523
|
| 3 |
+
size 19989325
|
tokenizer_config.json
CHANGED
|
@@ -1,239 +1,32 @@
|
|
| 1 |
{
|
| 2 |
-
"add_bos_token": false,
|
| 3 |
"add_prefix_space": false,
|
| 4 |
-
"
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
"normalized": false,
|
| 9 |
-
"rstrip": false,
|
| 10 |
-
"single_word": false,
|
| 11 |
-
"special": true
|
| 12 |
-
},
|
| 13 |
-
"151644": {
|
| 14 |
-
"content": "<|im_start|>",
|
| 15 |
-
"lstrip": false,
|
| 16 |
-
"normalized": false,
|
| 17 |
-
"rstrip": false,
|
| 18 |
-
"single_word": false,
|
| 19 |
-
"special": true
|
| 20 |
-
},
|
| 21 |
-
"151645": {
|
| 22 |
-
"content": "<|im_end|>",
|
| 23 |
-
"lstrip": false,
|
| 24 |
-
"normalized": false,
|
| 25 |
-
"rstrip": false,
|
| 26 |
-
"single_word": false,
|
| 27 |
-
"special": true
|
| 28 |
-
},
|
| 29 |
-
"151646": {
|
| 30 |
-
"content": "<|object_ref_start|>",
|
| 31 |
-
"lstrip": false,
|
| 32 |
-
"normalized": false,
|
| 33 |
-
"rstrip": false,
|
| 34 |
-
"single_word": false,
|
| 35 |
-
"special": true
|
| 36 |
-
},
|
| 37 |
-
"151647": {
|
| 38 |
-
"content": "<|object_ref_end|>",
|
| 39 |
-
"lstrip": false,
|
| 40 |
-
"normalized": false,
|
| 41 |
-
"rstrip": false,
|
| 42 |
-
"single_word": false,
|
| 43 |
-
"special": true
|
| 44 |
-
},
|
| 45 |
-
"151648": {
|
| 46 |
-
"content": "<|box_start|>",
|
| 47 |
-
"lstrip": false,
|
| 48 |
-
"normalized": false,
|
| 49 |
-
"rstrip": false,
|
| 50 |
-
"single_word": false,
|
| 51 |
-
"special": true
|
| 52 |
-
},
|
| 53 |
-
"151649": {
|
| 54 |
-
"content": "<|box_end|>",
|
| 55 |
-
"lstrip": false,
|
| 56 |
-
"normalized": false,
|
| 57 |
-
"rstrip": false,
|
| 58 |
-
"single_word": false,
|
| 59 |
-
"special": true
|
| 60 |
-
},
|
| 61 |
-
"151650": {
|
| 62 |
-
"content": "<|quad_start|>",
|
| 63 |
-
"lstrip": false,
|
| 64 |
-
"normalized": false,
|
| 65 |
-
"rstrip": false,
|
| 66 |
-
"single_word": false,
|
| 67 |
-
"special": true
|
| 68 |
-
},
|
| 69 |
-
"151651": {
|
| 70 |
-
"content": "<|quad_end|>",
|
| 71 |
-
"lstrip": false,
|
| 72 |
-
"normalized": false,
|
| 73 |
-
"rstrip": false,
|
| 74 |
-
"single_word": false,
|
| 75 |
-
"special": true
|
| 76 |
-
},
|
| 77 |
-
"151652": {
|
| 78 |
-
"content": "<|vision_start|>",
|
| 79 |
-
"lstrip": false,
|
| 80 |
-
"normalized": false,
|
| 81 |
-
"rstrip": false,
|
| 82 |
-
"single_word": false,
|
| 83 |
-
"special": true
|
| 84 |
-
},
|
| 85 |
-
"151653": {
|
| 86 |
-
"content": "<|vision_end|>",
|
| 87 |
-
"lstrip": false,
|
| 88 |
-
"normalized": false,
|
| 89 |
-
"rstrip": false,
|
| 90 |
-
"single_word": false,
|
| 91 |
-
"special": true
|
| 92 |
-
},
|
| 93 |
-
"151654": {
|
| 94 |
-
"content": "<|vision_pad|>",
|
| 95 |
-
"lstrip": false,
|
| 96 |
-
"normalized": false,
|
| 97 |
-
"rstrip": false,
|
| 98 |
-
"single_word": false,
|
| 99 |
-
"special": true
|
| 100 |
-
},
|
| 101 |
-
"151655": {
|
| 102 |
-
"content": "<|image_pad|>",
|
| 103 |
-
"lstrip": false,
|
| 104 |
-
"normalized": false,
|
| 105 |
-
"rstrip": false,
|
| 106 |
-
"single_word": false,
|
| 107 |
-
"special": true
|
| 108 |
-
},
|
| 109 |
-
"151656": {
|
| 110 |
-
"content": "<|video_pad|>",
|
| 111 |
-
"lstrip": false,
|
| 112 |
-
"normalized": false,
|
| 113 |
-
"rstrip": false,
|
| 114 |
-
"single_word": false,
|
| 115 |
-
"special": true
|
| 116 |
-
},
|
| 117 |
-
"151657": {
|
| 118 |
-
"content": "<tool_call>",
|
| 119 |
-
"lstrip": false,
|
| 120 |
-
"normalized": false,
|
| 121 |
-
"rstrip": false,
|
| 122 |
-
"single_word": false,
|
| 123 |
-
"special": false
|
| 124 |
-
},
|
| 125 |
-
"151658": {
|
| 126 |
-
"content": "</tool_call>",
|
| 127 |
-
"lstrip": false,
|
| 128 |
-
"normalized": false,
|
| 129 |
-
"rstrip": false,
|
| 130 |
-
"single_word": false,
|
| 131 |
-
"special": false
|
| 132 |
-
},
|
| 133 |
-
"151659": {
|
| 134 |
-
"content": "<|fim_prefix|>",
|
| 135 |
-
"lstrip": false,
|
| 136 |
-
"normalized": false,
|
| 137 |
-
"rstrip": false,
|
| 138 |
-
"single_word": false,
|
| 139 |
-
"special": false
|
| 140 |
-
},
|
| 141 |
-
"151660": {
|
| 142 |
-
"content": "<|fim_middle|>",
|
| 143 |
-
"lstrip": false,
|
| 144 |
-
"normalized": false,
|
| 145 |
-
"rstrip": false,
|
| 146 |
-
"single_word": false,
|
| 147 |
-
"special": false
|
| 148 |
-
},
|
| 149 |
-
"151661": {
|
| 150 |
-
"content": "<|fim_suffix|>",
|
| 151 |
-
"lstrip": false,
|
| 152 |
-
"normalized": false,
|
| 153 |
-
"rstrip": false,
|
| 154 |
-
"single_word": false,
|
| 155 |
-
"special": false
|
| 156 |
-
},
|
| 157 |
-
"151662": {
|
| 158 |
-
"content": "<|fim_pad|>",
|
| 159 |
-
"lstrip": false,
|
| 160 |
-
"normalized": false,
|
| 161 |
-
"rstrip": false,
|
| 162 |
-
"single_word": false,
|
| 163 |
-
"special": false
|
| 164 |
-
},
|
| 165 |
-
"151663": {
|
| 166 |
-
"content": "<|repo_name|>",
|
| 167 |
-
"lstrip": false,
|
| 168 |
-
"normalized": false,
|
| 169 |
-
"rstrip": false,
|
| 170 |
-
"single_word": false,
|
| 171 |
-
"special": false
|
| 172 |
-
},
|
| 173 |
-
"151664": {
|
| 174 |
-
"content": "<|file_sep|>",
|
| 175 |
-
"lstrip": false,
|
| 176 |
-
"normalized": false,
|
| 177 |
-
"rstrip": false,
|
| 178 |
-
"single_word": false,
|
| 179 |
-
"special": false
|
| 180 |
-
},
|
| 181 |
-
"151665": {
|
| 182 |
-
"content": "<tool_response>",
|
| 183 |
-
"lstrip": false,
|
| 184 |
-
"normalized": false,
|
| 185 |
-
"rstrip": false,
|
| 186 |
-
"single_word": false,
|
| 187 |
-
"special": false
|
| 188 |
-
},
|
| 189 |
-
"151666": {
|
| 190 |
-
"content": "</tool_response>",
|
| 191 |
-
"lstrip": false,
|
| 192 |
-
"normalized": false,
|
| 193 |
-
"rstrip": false,
|
| 194 |
-
"single_word": false,
|
| 195 |
-
"special": false
|
| 196 |
-
},
|
| 197 |
-
"151667": {
|
| 198 |
-
"content": "<think>",
|
| 199 |
-
"lstrip": false,
|
| 200 |
-
"normalized": false,
|
| 201 |
-
"rstrip": false,
|
| 202 |
-
"single_word": false,
|
| 203 |
-
"special": false
|
| 204 |
-
},
|
| 205 |
-
"151668": {
|
| 206 |
-
"content": "</think>",
|
| 207 |
-
"lstrip": false,
|
| 208 |
-
"normalized": false,
|
| 209 |
-
"rstrip": false,
|
| 210 |
-
"single_word": false,
|
| 211 |
-
"special": false
|
| 212 |
-
}
|
| 213 |
-
},
|
| 214 |
-
"additional_special_tokens": [
|
| 215 |
-
"<|im_start|>",
|
| 216 |
-
"<|im_end|>",
|
| 217 |
-
"<|object_ref_start|>",
|
| 218 |
-
"<|object_ref_end|>",
|
| 219 |
-
"<|box_start|>",
|
| 220 |
-
"<|box_end|>",
|
| 221 |
-
"<|quad_start|>",
|
| 222 |
-
"<|quad_end|>",
|
| 223 |
-
"<|vision_start|>",
|
| 224 |
-
"<|vision_end|>",
|
| 225 |
-
"<|vision_pad|>",
|
| 226 |
-
"<|image_pad|>",
|
| 227 |
-
"<|video_pad|>"
|
| 228 |
-
],
|
| 229 |
"bos_token": null,
|
| 230 |
"clean_up_tokenization_spaces": false,
|
| 231 |
"eos_token": "<|endoftext|>",
|
| 232 |
"errors": "replace",
|
| 233 |
-
"
|
| 234 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 235 |
"pad_token": "<|endoftext|>",
|
|
|
|
| 236 |
"split_special_tokens": false,
|
| 237 |
"tokenizer_class": "Qwen2Tokenizer",
|
| 238 |
-
"unk_token": null
|
|
|
|
|
|
|
|
|
|
| 239 |
}
|
|
|
|
| 1 |
{
|
|
|
|
| 2 |
"add_prefix_space": false,
|
| 3 |
+
"audio_bos_token": "<|audio_start|>",
|
| 4 |
+
"audio_eos_token": "<|audio_end|>",
|
| 5 |
+
"audio_token": "<|audio_pad|>",
|
| 6 |
+
"backend": "tokenizers",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7 |
"bos_token": null,
|
| 8 |
"clean_up_tokenization_spaces": false,
|
| 9 |
"eos_token": "<|endoftext|>",
|
| 10 |
"errors": "replace",
|
| 11 |
+
"image_token": "<|image_pad|>",
|
| 12 |
+
"is_local": false,
|
| 13 |
+
"local_files_only": false,
|
| 14 |
+
"model_max_length": 262144,
|
| 15 |
+
"model_specific_special_tokens": {
|
| 16 |
+
"audio_bos_token": "<|audio_start|>",
|
| 17 |
+
"audio_eos_token": "<|audio_end|>",
|
| 18 |
+
"audio_token": "<|audio_pad|>",
|
| 19 |
+
"image_token": "<|image_pad|>",
|
| 20 |
+
"video_token": "<|video_pad|>",
|
| 21 |
+
"vision_bos_token": "<|vision_start|>",
|
| 22 |
+
"vision_eos_token": "<|vision_end|>"
|
| 23 |
+
},
|
| 24 |
"pad_token": "<|endoftext|>",
|
| 25 |
+
"pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
|
| 26 |
"split_special_tokens": false,
|
| 27 |
"tokenizer_class": "Qwen2Tokenizer",
|
| 28 |
+
"unk_token": null,
|
| 29 |
+
"video_token": "<|video_pad|>",
|
| 30 |
+
"vision_bos_token": "<|vision_start|>",
|
| 31 |
+
"vision_eos_token": "<|vision_end|>"
|
| 32 |
}
|
train.log
CHANGED
|
@@ -1,317 +1,319 @@
|
|
| 1 |
-
|
|
|
|
| 2 |
12576 training requests (holdout=[]), questions by type {'score': 3448, 'noul': 5224, 'choice': 6904}
|
| 3 |
-
|
| 4 |
-
ep0 step
|
| 5 |
-
ep0 step
|
| 6 |
-
ep0 step
|
| 7 |
-
ep0 step
|
| 8 |
-
ep0 step
|
| 9 |
-
ep0 step
|
| 10 |
-
ep0 step
|
| 11 |
-
ep0 step
|
| 12 |
-
ep0 step
|
| 13 |
-
ep0 step
|
| 14 |
-
ep0 step
|
| 15 |
-
ep0 step
|
| 16 |
-
ep0 step
|
| 17 |
-
ep0 step
|
| 18 |
-
ep0 step
|
| 19 |
-
ep0 step
|
| 20 |
-
ep0 step
|
| 21 |
-
ep0 step
|
| 22 |
-
ep0 step
|
| 23 |
-
ep0 step
|
| 24 |
-
ep0 step
|
| 25 |
-
ep0 step
|
| 26 |
-
ep0 step
|
| 27 |
-
ep0 step
|
| 28 |
-
ep0 step
|
| 29 |
-
ep0 step
|
| 30 |
-
ep0 step
|
| 31 |
-
ep0 step
|
| 32 |
-
ep0 step
|
| 33 |
-
ep0 step
|
| 34 |
-
ep0 step
|
| 35 |
-
ep0 step
|
| 36 |
-
ep0 step
|
| 37 |
-
ep0 step
|
| 38 |
-
ep0 step
|
| 39 |
-
ep0 step
|
| 40 |
-
ep0 step
|
| 41 |
-
ep0 step
|
| 42 |
-
ep0 step
|
| 43 |
-
ep0 step
|
| 44 |
-
ep0 step
|
| 45 |
-
ep0 step
|
| 46 |
-
ep0 step
|
| 47 |
-
ep0 step
|
| 48 |
-
ep0 step
|
| 49 |
-
ep0 step
|
| 50 |
-
ep0 step
|
| 51 |
-
ep0 step
|
| 52 |
-
ep0 step
|
| 53 |
-
ep0 step
|
| 54 |
-
ep0 step
|
| 55 |
-
ep0 step
|
| 56 |
-
ep0 step
|
| 57 |
-
ep0 step
|
| 58 |
-
ep0 step
|
| 59 |
-
ep0 step
|
| 60 |
-
ep0 step
|
| 61 |
-
ep0 step
|
| 62 |
-
ep0 step
|
| 63 |
-
ep0 step
|
| 64 |
-
ep0 step
|
| 65 |
-
ep0 step
|
| 66 |
-
ep0 step
|
| 67 |
-
ep0 step
|
| 68 |
-
ep0 step
|
| 69 |
-
ep0 step
|
| 70 |
-
ep0 step
|
| 71 |
-
ep0 step
|
| 72 |
-
ep0 step
|
| 73 |
-
ep0 step
|
| 74 |
-
ep0 step
|
| 75 |
-
ep0 step
|
| 76 |
-
ep0 step
|
| 77 |
-
ep0 step
|
| 78 |
-
ep0 step
|
| 79 |
-
ep0 step
|
| 80 |
-
ep0 step
|
| 81 |
-
ep0 step
|
| 82 |
-
ep0 step
|
| 83 |
-
ep0 step
|
| 84 |
-
ep0 step
|
| 85 |
-
ep0 step
|
| 86 |
-
ep0 step
|
| 87 |
-
ep0 step
|
| 88 |
-
ep0 step
|
| 89 |
-
ep0 step
|
| 90 |
-
ep0 step
|
| 91 |
-
ep0 step
|
| 92 |
-
ep0 step
|
| 93 |
-
ep0 step
|
| 94 |
-
ep0 step
|
| 95 |
-
ep0 step
|
| 96 |
-
ep0 step
|
| 97 |
-
ep0 step
|
| 98 |
-
ep0 step
|
| 99 |
-
ep0 step
|
| 100 |
-
ep0 step
|
| 101 |
-
ep0 step
|
| 102 |
-
ep0 step
|
| 103 |
-
ep0 step
|
| 104 |
-
ep0 step
|
| 105 |
-
ep0 step
|
| 106 |
-
ep0 step
|
| 107 |
-
ep0 step
|
| 108 |
-
ep0 step
|
| 109 |
-
ep0 step
|
| 110 |
-
ep0 step
|
| 111 |
-
ep0 step
|
| 112 |
-
ep0 step
|
| 113 |
-
ep0 step
|
| 114 |
-
ep0 step
|
| 115 |
-
ep0 step
|
| 116 |
-
ep0 step
|
| 117 |
-
ep0 step
|
| 118 |
-
ep0 step
|
| 119 |
-
ep0 step
|
| 120 |
-
ep0 step
|
| 121 |
-
ep0 step
|
| 122 |
-
ep0 step
|
| 123 |
-
ep0 step
|
| 124 |
-
ep0 step
|
| 125 |
-
ep0 step
|
| 126 |
-
ep0 step
|
| 127 |
-
ep0 step
|
| 128 |
-
ep0 step
|
| 129 |
-
ep0 step
|
| 130 |
-
ep0 step
|
| 131 |
-
ep0 step
|
| 132 |
-
ep0 step
|
| 133 |
-
ep0 step
|
| 134 |
-
ep0 step
|
| 135 |
-
ep0 step
|
| 136 |
-
ep0 step
|
| 137 |
-
ep0 step
|
| 138 |
-
ep0 step
|
| 139 |
-
ep0 step
|
| 140 |
-
ep0 step
|
| 141 |
-
ep0 step
|
| 142 |
-
ep0 step
|
| 143 |
-
ep0 step
|
| 144 |
-
ep0 step
|
| 145 |
-
ep0 step
|
| 146 |
-
ep0 step
|
| 147 |
-
ep0 step
|
| 148 |
-
ep0 step
|
| 149 |
-
ep0 step
|
| 150 |
-
ep0 step
|
| 151 |
-
ep0 step
|
| 152 |
-
ep0 step
|
| 153 |
-
ep0 step
|
| 154 |
-
ep0 step
|
| 155 |
-
ep0 step
|
| 156 |
-
ep0 step
|
| 157 |
-
ep0 step
|
| 158 |
-
ep0 step
|
| 159 |
-
ep0 step
|
| 160 |
-
|
| 161 |
-
ep1 step
|
| 162 |
-
ep1 step
|
| 163 |
-
ep1 step
|
| 164 |
-
ep1 step
|
| 165 |
-
ep1 step
|
| 166 |
-
ep1 step
|
| 167 |
-
ep1 step
|
| 168 |
-
ep1 step
|
| 169 |
-
ep1 step
|
| 170 |
-
ep1 step
|
| 171 |
-
ep1 step
|
| 172 |
-
ep1 step
|
| 173 |
-
ep1 step
|
| 174 |
-
ep1 step
|
| 175 |
-
ep1 step
|
| 176 |
-
ep1 step
|
| 177 |
-
ep1 step
|
| 178 |
-
ep1 step
|
| 179 |
-
ep1 step
|
| 180 |
-
ep1 step
|
| 181 |
-
ep1 step
|
| 182 |
-
ep1 step
|
| 183 |
-
ep1 step
|
| 184 |
-
ep1 step
|
| 185 |
-
ep1 step
|
| 186 |
-
ep1 step
|
| 187 |
-
ep1 step
|
| 188 |
-
ep1 step
|
| 189 |
-
ep1 step
|
| 190 |
-
ep1 step
|
| 191 |
-
ep1 step
|
| 192 |
-
ep1 step
|
| 193 |
-
ep1 step
|
| 194 |
-
ep1 step
|
| 195 |
-
ep1 step
|
| 196 |
-
ep1 step
|
| 197 |
-
ep1 step
|
| 198 |
-
ep1 step
|
| 199 |
-
ep1 step
|
| 200 |
-
ep1 step
|
| 201 |
-
ep1 step
|
| 202 |
-
ep1 step
|
| 203 |
-
ep1 step
|
| 204 |
-
ep1 step
|
| 205 |
-
ep1 step
|
| 206 |
-
ep1 step
|
| 207 |
-
ep1 step
|
| 208 |
-
ep1 step
|
| 209 |
-
ep1 step
|
| 210 |
-
ep1 step
|
| 211 |
-
ep1 step
|
| 212 |
-
ep1 step
|
| 213 |
-
ep1 step
|
| 214 |
-
ep1 step
|
| 215 |
-
ep1 step
|
| 216 |
-
ep1 step
|
| 217 |
-
ep1 step
|
| 218 |
-
ep1 step
|
| 219 |
-
ep1 step
|
| 220 |
-
ep1 step
|
| 221 |
-
ep1 step
|
| 222 |
-
ep1 step
|
| 223 |
-
ep1 step
|
| 224 |
-
ep1 step
|
| 225 |
-
ep1 step
|
| 226 |
-
ep1 step
|
| 227 |
-
ep1 step
|
| 228 |
-
ep1 step
|
| 229 |
-
ep1 step
|
| 230 |
-
ep1 step
|
| 231 |
-
ep1 step
|
| 232 |
-
ep1 step
|
| 233 |
-
ep1 step
|
| 234 |
-
ep1 step
|
| 235 |
-
ep1 step
|
| 236 |
-
ep1 step
|
| 237 |
-
ep1 step
|
| 238 |
-
ep1 step
|
| 239 |
-
ep1 step
|
| 240 |
-
ep1 step
|
| 241 |
-
ep1 step
|
| 242 |
-
ep1 step
|
| 243 |
-
ep1 step
|
| 244 |
-
ep1 step
|
| 245 |
-
ep1 step
|
| 246 |
-
ep1 step
|
| 247 |
-
ep1 step
|
| 248 |
-
ep1 step
|
| 249 |
-
ep1 step
|
| 250 |
-
ep1 step
|
| 251 |
-
ep1 step
|
| 252 |
-
ep1 step
|
| 253 |
-
ep1 step
|
| 254 |
-
ep1 step
|
| 255 |
-
ep1 step
|
| 256 |
-
ep1 step
|
| 257 |
-
ep1 step
|
| 258 |
-
ep1 step
|
| 259 |
-
ep1 step
|
| 260 |
-
ep1 step
|
| 261 |
-
ep1 step
|
| 262 |
-
ep1 step
|
| 263 |
-
ep1 step
|
| 264 |
-
ep1 step
|
| 265 |
-
ep1 step
|
| 266 |
-
ep1 step
|
| 267 |
-
ep1 step
|
| 268 |
-
ep1 step
|
| 269 |
-
ep1 step
|
| 270 |
-
ep1 step
|
| 271 |
-
ep1 step
|
| 272 |
-
ep1 step
|
| 273 |
-
ep1 step
|
| 274 |
-
ep1 step
|
| 275 |
-
ep1 step
|
| 276 |
-
ep1 step
|
| 277 |
-
ep1 step
|
| 278 |
-
ep1 step
|
| 279 |
-
ep1 step
|
| 280 |
-
ep1 step
|
| 281 |
-
ep1 step
|
| 282 |
-
ep1 step
|
| 283 |
-
ep1 step
|
| 284 |
-
ep1 step
|
| 285 |
-
ep1 step
|
| 286 |
-
ep1 step
|
| 287 |
-
ep1 step
|
| 288 |
-
ep1 step
|
| 289 |
-
ep1 step
|
| 290 |
-
ep1 step
|
| 291 |
-
ep1 step
|
| 292 |
-
ep1 step
|
| 293 |
-
ep1 step
|
| 294 |
-
ep1 step
|
| 295 |
-
ep1 step
|
| 296 |
-
ep1 step
|
| 297 |
-
ep1 step
|
| 298 |
-
ep1 step
|
| 299 |
-
ep1 step
|
| 300 |
-
ep1 step
|
| 301 |
-
ep1 step
|
| 302 |
-
ep1 step
|
| 303 |
-
ep1 step
|
| 304 |
-
ep1 step
|
| 305 |
-
ep1 step
|
| 306 |
-
ep1 step
|
| 307 |
-
ep1 step
|
| 308 |
-
ep1 step
|
| 309 |
-
ep1 step
|
| 310 |
-
ep1 step
|
| 311 |
-
ep1 step
|
| 312 |
-
ep1 step
|
| 313 |
-
ep1 step
|
| 314 |
-
ep1 step
|
| 315 |
-
ep1 step
|
| 316 |
-
ep1 step
|
| 317 |
-
|
|
|
|
|
|
| 1 |
+
Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
|
| 2 |
+
device=cuda trainable params=33.8M
|
| 3 |
12576 training requests (holdout=[]), questions by type {'score': 3448, 'noul': 5224, 'choice': 6904}
|
| 4 |
+
[transformers] `causal_conv1d_fn` is falling back to its reference PyTorch implementation because `causal_conv1d` is not installed. This is correct but much slower; install `causal_conv1d` for the optimized kernel.
|
| 5 |
+
ep0 step 10/3144 loss 2.066 kl 0.000 anchor 0.000 1.741s/rec
|
| 6 |
+
ep0 step 20/3144 loss 1.849 kl 0.000 anchor 0.000 0.938s/rec
|
| 7 |
+
ep0 step 30/3144 loss 1.933 kl 0.000 anchor 0.000 0.652s/rec
|
| 8 |
+
ep0 step 40/3144 loss 1.695 kl 0.000 anchor 0.000 0.513s/rec
|
| 9 |
+
ep0 step 50/3144 loss 1.560 kl 0.000 anchor 0.000 0.480s/rec
|
| 10 |
+
ep0 step 60/3144 loss 1.408 kl 0.000 anchor 0.000 0.418s/rec
|
| 11 |
+
ep0 step 70/3144 loss 1.757 kl 0.000 anchor 0.000 0.377s/rec
|
| 12 |
+
ep0 step 80/3144 loss 1.322 kl 0.000 anchor 0.000 0.368s/rec
|
| 13 |
+
ep0 step 90/3144 loss 1.148 kl 0.000 anchor 0.000 0.335s/rec
|
| 14 |
+
ep0 step 100/3144 loss 0.859 kl 0.000 anchor 0.000 0.312s/rec
|
| 15 |
+
ep0 step 110/3144 loss 0.970 kl 0.000 anchor 0.000 0.291s/rec
|
| 16 |
+
ep0 step 120/3144 loss 0.870 kl 0.000 anchor 0.000 0.275s/rec
|
| 17 |
+
ep0 step 130/3144 loss 0.980 kl 0.000 anchor 0.000 0.261s/rec
|
| 18 |
+
ep0 step 140/3144 loss 0.758 kl 0.000 anchor 0.000 0.249s/rec
|
| 19 |
+
ep0 step 150/3144 loss 0.843 kl 0.000 anchor 0.000 0.241s/rec
|
| 20 |
+
ep0 step 160/3144 loss 0.771 kl 0.000 anchor 0.000 0.233s/rec
|
| 21 |
+
ep0 step 170/3144 loss 0.936 kl 0.000 anchor 0.000 0.226s/rec
|
| 22 |
+
ep0 step 180/3144 loss 0.669 kl 0.000 anchor 0.000 0.219s/rec
|
| 23 |
+
ep0 step 190/3144 loss 0.883 kl 0.000 anchor 0.000 0.215s/rec
|
| 24 |
+
ep0 step 200/3144 loss 0.600 kl 0.000 anchor 0.000 0.210s/rec
|
| 25 |
+
ep0 step 210/3144 loss 0.918 kl 0.000 anchor 0.000 0.206s/rec
|
| 26 |
+
ep0 step 220/3144 loss 0.831 kl 0.000 anchor 0.000 0.202s/rec
|
| 27 |
+
ep0 step 230/3144 loss 0.598 kl 0.000 anchor 0.000 0.198s/rec
|
| 28 |
+
ep0 step 240/3144 loss 0.652 kl 0.000 anchor 0.000 0.194s/rec
|
| 29 |
+
ep0 step 250/3144 loss 0.798 kl 0.000 anchor 0.000 0.191s/rec
|
| 30 |
+
ep0 step 260/3144 loss 0.616 kl 0.000 anchor 0.000 0.187s/rec
|
| 31 |
+
ep0 step 270/3144 loss 0.546 kl 0.000 anchor 0.000 0.184s/rec
|
| 32 |
+
ep0 step 280/3144 loss 0.622 kl 0.000 anchor 0.000 0.182s/rec
|
| 33 |
+
ep0 step 290/3144 loss 0.627 kl 0.000 anchor 0.000 0.180s/rec
|
| 34 |
+
ep0 step 300/3144 loss 0.729 kl 0.000 anchor 0.000 0.177s/rec
|
| 35 |
+
ep0 step 310/3144 loss 0.612 kl 0.000 anchor 0.000 0.175s/rec
|
| 36 |
+
ep0 step 320/3144 loss 0.557 kl 0.000 anchor 0.000 0.173s/rec
|
| 37 |
+
ep0 step 330/3144 loss 0.520 kl 0.000 anchor 0.000 0.171s/rec
|
| 38 |
+
ep0 step 340/3144 loss 0.592 kl 0.000 anchor 0.000 0.169s/rec
|
| 39 |
+
ep0 step 350/3144 loss 0.665 kl 0.000 anchor 0.000 0.167s/rec
|
| 40 |
+
ep0 step 360/3144 loss 0.408 kl 0.000 anchor 0.000 0.166s/rec
|
| 41 |
+
ep0 step 370/3144 loss 0.717 kl 0.000 anchor 0.000 0.164s/rec
|
| 42 |
+
ep0 step 380/3144 loss 0.537 kl 0.000 anchor 0.000 0.162s/rec
|
| 43 |
+
ep0 step 390/3144 loss 0.735 kl 0.000 anchor 0.000 0.161s/rec
|
| 44 |
+
ep0 step 400/3144 loss 0.982 kl 0.000 anchor 0.000 0.160s/rec
|
| 45 |
+
ep0 step 410/3144 loss 0.610 kl 0.000 anchor 0.000 0.158s/rec
|
| 46 |
+
ep0 step 420/3144 loss 0.420 kl 0.000 anchor 0.000 0.157s/rec
|
| 47 |
+
ep0 step 430/3144 loss 0.661 kl 0.000 anchor 0.000 0.156s/rec
|
| 48 |
+
ep0 step 440/3144 loss 0.834 kl 0.000 anchor 0.000 0.154s/rec
|
| 49 |
+
ep0 step 450/3144 loss 0.546 kl 0.000 anchor 0.000 0.153s/rec
|
| 50 |
+
ep0 step 460/3144 loss 0.533 kl 0.000 anchor 0.000 0.152s/rec
|
| 51 |
+
ep0 step 470/3144 loss 0.523 kl 0.000 anchor 0.000 0.151s/rec
|
| 52 |
+
ep0 step 480/3144 loss 0.490 kl 0.000 anchor 0.000 0.150s/rec
|
| 53 |
+
ep0 step 490/3144 loss 0.412 kl 0.000 anchor 0.000 0.149s/rec
|
| 54 |
+
ep0 step 500/3144 loss 0.536 kl 0.000 anchor 0.000 0.148s/rec
|
| 55 |
+
ep0 step 510/3144 loss 0.509 kl 0.000 anchor 0.000 0.147s/rec
|
| 56 |
+
ep0 step 520/3144 loss 0.453 kl 0.000 anchor 0.000 0.147s/rec
|
| 57 |
+
ep0 step 530/3144 loss 0.932 kl 0.000 anchor 0.000 0.146s/rec
|
| 58 |
+
ep0 step 540/3144 loss 0.465 kl 0.000 anchor 0.000 0.145s/rec
|
| 59 |
+
ep0 step 550/3144 loss 0.358 kl 0.000 anchor 0.000 0.144s/rec
|
| 60 |
+
ep0 step 560/3144 loss 0.578 kl 0.000 anchor 0.000 0.144s/rec
|
| 61 |
+
ep0 step 570/3144 loss 0.565 kl 0.000 anchor 0.000 0.143s/rec
|
| 62 |
+
ep0 step 580/3144 loss 0.429 kl 0.000 anchor 0.000 0.143s/rec
|
| 63 |
+
ep0 step 590/3144 loss 0.475 kl 0.000 anchor 0.000 0.142s/rec
|
| 64 |
+
ep0 step 600/3144 loss 0.729 kl 0.000 anchor 0.000 0.142s/rec
|
| 65 |
+
ep0 step 610/3144 loss 0.368 kl 0.000 anchor 0.000 0.141s/rec
|
| 66 |
+
ep0 step 620/3144 loss 0.404 kl 0.000 anchor 0.000 0.140s/rec
|
| 67 |
+
ep0 step 630/3144 loss 0.602 kl 0.000 anchor 0.000 0.140s/rec
|
| 68 |
+
ep0 step 640/3144 loss 0.676 kl 0.000 anchor 0.000 0.140s/rec
|
| 69 |
+
ep0 step 650/3144 loss 0.863 kl 0.000 anchor 0.000 0.139s/rec
|
| 70 |
+
ep0 step 660/3144 loss 0.709 kl 0.000 anchor 0.000 0.139s/rec
|
| 71 |
+
ep0 step 670/3144 loss 0.692 kl 0.000 anchor 0.000 0.138s/rec
|
| 72 |
+
ep0 step 680/3144 loss 0.635 kl 0.000 anchor 0.000 0.137s/rec
|
| 73 |
+
ep0 step 690/3144 loss 0.473 kl 0.000 anchor 0.000 0.137s/rec
|
| 74 |
+
ep0 step 700/3144 loss 0.507 kl 0.000 anchor 0.000 0.136s/rec
|
| 75 |
+
ep0 step 710/3144 loss 0.487 kl 0.000 anchor 0.000 0.136s/rec
|
| 76 |
+
ep0 step 720/3144 loss 0.247 kl 0.000 anchor 0.000 0.135s/rec
|
| 77 |
+
ep0 step 730/3144 loss 0.578 kl 0.000 anchor 0.000 0.135s/rec
|
| 78 |
+
ep0 step 740/3144 loss 0.446 kl 0.000 anchor 0.000 0.135s/rec
|
| 79 |
+
ep0 step 750/3144 loss 0.732 kl 0.000 anchor 0.000 0.134s/rec
|
| 80 |
+
ep0 step 760/3144 loss 0.308 kl 0.000 anchor 0.000 0.134s/rec
|
| 81 |
+
ep0 step 770/3144 loss 0.373 kl 0.000 anchor 0.000 0.133s/rec
|
| 82 |
+
ep0 step 780/3144 loss 0.794 kl 0.000 anchor 0.000 0.133s/rec
|
| 83 |
+
ep0 step 790/3144 loss 0.372 kl 0.000 anchor 0.000 0.132s/rec
|
| 84 |
+
ep0 step 800/3144 loss 0.632 kl 0.000 anchor 0.000 0.132s/rec
|
| 85 |
+
ep0 step 810/3144 loss 0.487 kl 0.000 anchor 0.000 0.131s/rec
|
| 86 |
+
ep0 step 820/3144 loss 0.355 kl 0.000 anchor 0.000 0.131s/rec
|
| 87 |
+
ep0 step 830/3144 loss 0.460 kl 0.000 anchor 0.000 0.131s/rec
|
| 88 |
+
ep0 step 840/3144 loss 0.531 kl 0.000 anchor 0.000 0.130s/rec
|
| 89 |
+
ep0 step 850/3144 loss 0.575 kl 0.000 anchor 0.000 0.130s/rec
|
| 90 |
+
ep0 step 860/3144 loss 0.363 kl 0.000 anchor 0.000 0.130s/rec
|
| 91 |
+
ep0 step 870/3144 loss 0.708 kl 0.000 anchor 0.000 0.129s/rec
|
| 92 |
+
ep0 step 880/3144 loss 0.742 kl 0.000 anchor 0.000 0.129s/rec
|
| 93 |
+
ep0 step 890/3144 loss 0.415 kl 0.000 anchor 0.000 0.129s/rec
|
| 94 |
+
ep0 step 900/3144 loss 0.530 kl 0.000 anchor 0.000 0.129s/rec
|
| 95 |
+
ep0 step 910/3144 loss 0.543 kl 0.000 anchor 0.000 0.128s/rec
|
| 96 |
+
ep0 step 920/3144 loss 0.383 kl 0.000 anchor 0.000 0.128s/rec
|
| 97 |
+
ep0 step 930/3144 loss 0.381 kl 0.000 anchor 0.000 0.128s/rec
|
| 98 |
+
ep0 step 940/3144 loss 0.642 kl 0.000 anchor 0.000 0.128s/rec
|
| 99 |
+
ep0 step 950/3144 loss 0.260 kl 0.000 anchor 0.000 0.127s/rec
|
| 100 |
+
ep0 step 960/3144 loss 1.169 kl 0.000 anchor 0.000 0.127s/rec
|
| 101 |
+
ep0 step 970/3144 loss 0.387 kl 0.000 anchor 0.000 0.127s/rec
|
| 102 |
+
ep0 step 980/3144 loss 0.278 kl 0.000 anchor 0.000 0.127s/rec
|
| 103 |
+
ep0 step 990/3144 loss 0.393 kl 0.000 anchor 0.000 0.127s/rec
|
| 104 |
+
ep0 step 1000/3144 loss 0.501 kl 0.000 anchor 0.000 0.127s/rec
|
| 105 |
+
ep0 step 1010/3144 loss 0.529 kl 0.000 anchor 0.000 0.126s/rec
|
| 106 |
+
ep0 step 1020/3144 loss 0.606 kl 0.000 anchor 0.000 0.126s/rec
|
| 107 |
+
ep0 step 1030/3144 loss 0.436 kl 0.000 anchor 0.000 0.126s/rec
|
| 108 |
+
ep0 step 1040/3144 loss 0.454 kl 0.000 anchor 0.000 0.126s/rec
|
| 109 |
+
ep0 step 1050/3144 loss 0.346 kl 0.000 anchor 0.000 0.126s/rec
|
| 110 |
+
ep0 step 1060/3144 loss 0.434 kl 0.000 anchor 0.000 0.126s/rec
|
| 111 |
+
ep0 step 1070/3144 loss 0.528 kl 0.000 anchor 0.000 0.125s/rec
|
| 112 |
+
ep0 step 1080/3144 loss 0.392 kl 0.000 anchor 0.000 0.125s/rec
|
| 113 |
+
ep0 step 1090/3144 loss 0.626 kl 0.000 anchor 0.000 0.125s/rec
|
| 114 |
+
ep0 step 1100/3144 loss 0.337 kl 0.000 anchor 0.000 0.125s/rec
|
| 115 |
+
ep0 step 1110/3144 loss 0.354 kl 0.000 anchor 0.000 0.125s/rec
|
| 116 |
+
ep0 step 1120/3144 loss 0.401 kl 0.000 anchor 0.000 0.124s/rec
|
| 117 |
+
ep0 step 1130/3144 loss 0.565 kl 0.000 anchor 0.000 0.124s/rec
|
| 118 |
+
ep0 step 1140/3144 loss 0.295 kl 0.000 anchor 0.000 0.124s/rec
|
| 119 |
+
ep0 step 1150/3144 loss 0.319 kl 0.000 anchor 0.000 0.124s/rec
|
| 120 |
+
ep0 step 1160/3144 loss 0.551 kl 0.000 anchor 0.000 0.124s/rec
|
| 121 |
+
ep0 step 1170/3144 loss 0.440 kl 0.000 anchor 0.000 0.124s/rec
|
| 122 |
+
ep0 step 1180/3144 loss 0.386 kl 0.000 anchor 0.000 0.123s/rec
|
| 123 |
+
ep0 step 1190/3144 loss 0.452 kl 0.000 anchor 0.000 0.123s/rec
|
| 124 |
+
ep0 step 1200/3144 loss 0.478 kl 0.000 anchor 0.000 0.123s/rec
|
| 125 |
+
ep0 step 1210/3144 loss 0.400 kl 0.000 anchor 0.000 0.123s/rec
|
| 126 |
+
ep0 step 1220/3144 loss 0.502 kl 0.000 anchor 0.000 0.123s/rec
|
| 127 |
+
ep0 step 1230/3144 loss 0.458 kl 0.000 anchor 0.000 0.123s/rec
|
| 128 |
+
ep0 step 1240/3144 loss 0.431 kl 0.000 anchor 0.000 0.123s/rec
|
| 129 |
+
ep0 step 1250/3144 loss 0.528 kl 0.000 anchor 0.000 0.123s/rec
|
| 130 |
+
ep0 step 1260/3144 loss 0.375 kl 0.000 anchor 0.000 0.122s/rec
|
| 131 |
+
ep0 step 1270/3144 loss 0.492 kl 0.000 anchor 0.000 0.122s/rec
|
| 132 |
+
ep0 step 1280/3144 loss 0.443 kl 0.000 anchor 0.000 0.122s/rec
|
| 133 |
+
ep0 step 1290/3144 loss 0.487 kl 0.000 anchor 0.000 0.122s/rec
|
| 134 |
+
ep0 step 1300/3144 loss 0.714 kl 0.000 anchor 0.000 0.122s/rec
|
| 135 |
+
ep0 step 1310/3144 loss 0.445 kl 0.000 anchor 0.000 0.122s/rec
|
| 136 |
+
ep0 step 1320/3144 loss 0.481 kl 0.000 anchor 0.000 0.122s/rec
|
| 137 |
+
ep0 step 1330/3144 loss 0.695 kl 0.000 anchor 0.000 0.122s/rec
|
| 138 |
+
ep0 step 1340/3144 loss 0.527 kl 0.000 anchor 0.000 0.121s/rec
|
| 139 |
+
ep0 step 1350/3144 loss 0.310 kl 0.000 anchor 0.000 0.121s/rec
|
| 140 |
+
ep0 step 1360/3144 loss 0.617 kl 0.000 anchor 0.000 0.121s/rec
|
| 141 |
+
ep0 step 1370/3144 loss 0.565 kl 0.000 anchor 0.000 0.121s/rec
|
| 142 |
+
ep0 step 1380/3144 loss 0.547 kl 0.000 anchor 0.000 0.121s/rec
|
| 143 |
+
ep0 step 1390/3144 loss 0.360 kl 0.000 anchor 0.000 0.121s/rec
|
| 144 |
+
ep0 step 1400/3144 loss 0.686 kl 0.000 anchor 0.000 0.121s/rec
|
| 145 |
+
ep0 step 1410/3144 loss 0.337 kl 0.000 anchor 0.000 0.121s/rec
|
| 146 |
+
ep0 step 1420/3144 loss 0.455 kl 0.000 anchor 0.000 0.120s/rec
|
| 147 |
+
ep0 step 1430/3144 loss 0.393 kl 0.000 anchor 0.000 0.120s/rec
|
| 148 |
+
ep0 step 1440/3144 loss 0.452 kl 0.000 anchor 0.000 0.120s/rec
|
| 149 |
+
ep0 step 1450/3144 loss 0.484 kl 0.000 anchor 0.000 0.120s/rec
|
| 150 |
+
ep0 step 1460/3144 loss 0.624 kl 0.000 anchor 0.000 0.120s/rec
|
| 151 |
+
ep0 step 1470/3144 loss 0.600 kl 0.000 anchor 0.000 0.120s/rec
|
| 152 |
+
ep0 step 1480/3144 loss 0.420 kl 0.000 anchor 0.000 0.120s/rec
|
| 153 |
+
ep0 step 1490/3144 loss 0.359 kl 0.000 anchor 0.000 0.120s/rec
|
| 154 |
+
ep0 step 1500/3144 loss 0.469 kl 0.000 anchor 0.000 0.120s/rec
|
| 155 |
+
ep0 step 1510/3144 loss 0.365 kl 0.000 anchor 0.000 0.119s/rec
|
| 156 |
+
ep0 step 1520/3144 loss 0.476 kl 0.000 anchor 0.000 0.119s/rec
|
| 157 |
+
ep0 step 1530/3144 loss 0.374 kl 0.000 anchor 0.000 0.119s/rec
|
| 158 |
+
ep0 step 1540/3144 loss 0.332 kl 0.000 anchor 0.000 0.119s/rec
|
| 159 |
+
ep0 step 1550/3144 loss 0.354 kl 0.000 anchor 0.000 0.119s/rec
|
| 160 |
+
ep0 step 1560/3144 loss 0.226 kl 0.000 anchor 0.000 0.119s/rec
|
| 161 |
+
ep0 step 1570/3144 loss 0.675 kl 0.000 anchor 0.000 0.119s/rec
|
| 162 |
+
ep1 step 1580/3144 loss 0.710 kl 0.000 anchor 0.000 0.119s/rec
|
| 163 |
+
ep1 step 1590/3144 loss 0.157 kl 0.000 anchor 0.000 0.119s/rec
|
| 164 |
+
ep1 step 1600/3144 loss 0.564 kl 0.000 anchor 0.000 0.118s/rec
|
| 165 |
+
ep1 step 1610/3144 loss 0.229 kl 0.000 anchor 0.000 0.118s/rec
|
| 166 |
+
ep1 step 1620/3144 loss 0.446 kl 0.000 anchor 0.000 0.118s/rec
|
| 167 |
+
ep1 step 1630/3144 loss 0.189 kl 0.000 anchor 0.000 0.118s/rec
|
| 168 |
+
ep1 step 1640/3144 loss 0.166 kl 0.000 anchor 0.000 0.118s/rec
|
| 169 |
+
ep1 step 1650/3144 loss 0.437 kl 0.000 anchor 0.000 0.118s/rec
|
| 170 |
+
ep1 step 1660/3144 loss 0.439 kl 0.000 anchor 0.000 0.118s/rec
|
| 171 |
+
ep1 step 1670/3144 loss 0.361 kl 0.000 anchor 0.000 0.118s/rec
|
| 172 |
+
ep1 step 1680/3144 loss 0.296 kl 0.000 anchor 0.000 0.118s/rec
|
| 173 |
+
ep1 step 1690/3144 loss 0.229 kl 0.000 anchor 0.000 0.118s/rec
|
| 174 |
+
ep1 step 1700/3144 loss 0.453 kl 0.000 anchor 0.000 0.118s/rec
|
| 175 |
+
ep1 step 1710/3144 loss 0.279 kl 0.000 anchor 0.000 0.117s/rec
|
| 176 |
+
ep1 step 1720/3144 loss 0.937 kl 0.000 anchor 0.000 0.117s/rec
|
| 177 |
+
ep1 step 1730/3144 loss 0.396 kl 0.000 anchor 0.000 0.117s/rec
|
| 178 |
+
ep1 step 1740/3144 loss 0.152 kl 0.000 anchor 0.000 0.117s/rec
|
| 179 |
+
ep1 step 1750/3144 loss 0.231 kl 0.000 anchor 0.000 0.117s/rec
|
| 180 |
+
ep1 step 1760/3144 loss 0.347 kl 0.000 anchor 0.000 0.117s/rec
|
| 181 |
+
ep1 step 1770/3144 loss 0.296 kl 0.000 anchor 0.000 0.117s/rec
|
| 182 |
+
ep1 step 1780/3144 loss 0.330 kl 0.000 anchor 0.000 0.117s/rec
|
| 183 |
+
ep1 step 1790/3144 loss 0.320 kl 0.000 anchor 0.000 0.117s/rec
|
| 184 |
+
ep1 step 1800/3144 loss 0.356 kl 0.000 anchor 0.000 0.117s/rec
|
| 185 |
+
ep1 step 1810/3144 loss 0.311 kl 0.000 anchor 0.000 0.117s/rec
|
| 186 |
+
ep1 step 1820/3144 loss 0.167 kl 0.000 anchor 0.000 0.117s/rec
|
| 187 |
+
ep1 step 1830/3144 loss 0.402 kl 0.000 anchor 0.000 0.117s/rec
|
| 188 |
+
ep1 step 1840/3144 loss 0.335 kl 0.000 anchor 0.000 0.117s/rec
|
| 189 |
+
ep1 step 1850/3144 loss 0.322 kl 0.000 anchor 0.000 0.116s/rec
|
| 190 |
+
ep1 step 1860/3144 loss 0.659 kl 0.000 anchor 0.000 0.116s/rec
|
| 191 |
+
ep1 step 1870/3144 loss 0.214 kl 0.000 anchor 0.000 0.116s/rec
|
| 192 |
+
ep1 step 1880/3144 loss 0.202 kl 0.000 anchor 0.000 0.116s/rec
|
| 193 |
+
ep1 step 1890/3144 loss 0.322 kl 0.000 anchor 0.000 0.116s/rec
|
| 194 |
+
ep1 step 1900/3144 loss 0.302 kl 0.000 anchor 0.000 0.116s/rec
|
| 195 |
+
ep1 step 1910/3144 loss 0.203 kl 0.000 anchor 0.000 0.116s/rec
|
| 196 |
+
ep1 step 1920/3144 loss 0.352 kl 0.000 anchor 0.000 0.116s/rec
|
| 197 |
+
ep1 step 1930/3144 loss 0.390 kl 0.000 anchor 0.000 0.116s/rec
|
| 198 |
+
ep1 step 1940/3144 loss 0.308 kl 0.000 anchor 0.000 0.116s/rec
|
| 199 |
+
ep1 step 1950/3144 loss 0.682 kl 0.000 anchor 0.000 0.116s/rec
|
| 200 |
+
ep1 step 1960/3144 loss 0.190 kl 0.000 anchor 0.000 0.116s/rec
|
| 201 |
+
ep1 step 1970/3144 loss 0.330 kl 0.000 anchor 0.000 0.116s/rec
|
| 202 |
+
ep1 step 1980/3144 loss 0.274 kl 0.000 anchor 0.000 0.116s/rec
|
| 203 |
+
ep1 step 1990/3144 loss 0.178 kl 0.000 anchor 0.000 0.116s/rec
|
| 204 |
+
ep1 step 2000/3144 loss 0.186 kl 0.000 anchor 0.000 0.115s/rec
|
| 205 |
+
ep1 step 2010/3144 loss 0.425 kl 0.000 anchor 0.000 0.115s/rec
|
| 206 |
+
ep1 step 2020/3144 loss 0.413 kl 0.000 anchor 0.000 0.115s/rec
|
| 207 |
+
ep1 step 2030/3144 loss 0.519 kl 0.000 anchor 0.000 0.115s/rec
|
| 208 |
+
ep1 step 2040/3144 loss 0.163 kl 0.000 anchor 0.000 0.115s/rec
|
| 209 |
+
ep1 step 2050/3144 loss 0.106 kl 0.000 anchor 0.000 0.115s/rec
|
| 210 |
+
ep1 step 2060/3144 loss 0.301 kl 0.000 anchor 0.000 0.115s/rec
|
| 211 |
+
ep1 step 2070/3144 loss 0.208 kl 0.000 anchor 0.000 0.115s/rec
|
| 212 |
+
ep1 step 2080/3144 loss 0.386 kl 0.000 anchor 0.000 0.115s/rec
|
| 213 |
+
ep1 step 2090/3144 loss 0.218 kl 0.000 anchor 0.000 0.115s/rec
|
| 214 |
+
ep1 step 2100/3144 loss 0.408 kl 0.000 anchor 0.000 0.115s/rec
|
| 215 |
+
ep1 step 2110/3144 loss 0.357 kl 0.000 anchor 0.000 0.115s/rec
|
| 216 |
+
ep1 step 2120/3144 loss 0.354 kl 0.000 anchor 0.000 0.115s/rec
|
| 217 |
+
ep1 step 2130/3144 loss 0.381 kl 0.000 anchor 0.000 0.115s/rec
|
| 218 |
+
ep1 step 2140/3144 loss 0.263 kl 0.000 anchor 0.000 0.115s/rec
|
| 219 |
+
ep1 step 2150/3144 loss 0.668 kl 0.000 anchor 0.000 0.115s/rec
|
| 220 |
+
ep1 step 2160/3144 loss 0.267 kl 0.000 anchor 0.000 0.115s/rec
|
| 221 |
+
ep1 step 2170/3144 loss 0.110 kl 0.000 anchor 0.000 0.115s/rec
|
| 222 |
+
ep1 step 2180/3144 loss 0.288 kl 0.000 anchor 0.000 0.114s/rec
|
| 223 |
+
ep1 step 2190/3144 loss 0.257 kl 0.000 anchor 0.000 0.114s/rec
|
| 224 |
+
ep1 step 2200/3144 loss 0.316 kl 0.000 anchor 0.000 0.114s/rec
|
| 225 |
+
ep1 step 2210/3144 loss 0.139 kl 0.000 anchor 0.000 0.114s/rec
|
| 226 |
+
ep1 step 2220/3144 loss 0.356 kl 0.000 anchor 0.000 0.114s/rec
|
| 227 |
+
ep1 step 2230/3144 loss 0.331 kl 0.000 anchor 0.000 0.114s/rec
|
| 228 |
+
ep1 step 2240/3144 loss 0.163 kl 0.000 anchor 0.000 0.114s/rec
|
| 229 |
+
ep1 step 2250/3144 loss 0.141 kl 0.000 anchor 0.000 0.114s/rec
|
| 230 |
+
ep1 step 2260/3144 loss 0.175 kl 0.000 anchor 0.000 0.114s/rec
|
| 231 |
+
ep1 step 2270/3144 loss 0.383 kl 0.000 anchor 0.000 0.114s/rec
|
| 232 |
+
ep1 step 2280/3144 loss 0.245 kl 0.000 anchor 0.000 0.114s/rec
|
| 233 |
+
ep1 step 2290/3144 loss 0.166 kl 0.000 anchor 0.000 0.114s/rec
|
| 234 |
+
ep1 step 2300/3144 loss 0.245 kl 0.000 anchor 0.000 0.114s/rec
|
| 235 |
+
ep1 step 2310/3144 loss 0.328 kl 0.000 anchor 0.000 0.114s/rec
|
| 236 |
+
ep1 step 2320/3144 loss 0.196 kl 0.000 anchor 0.000 0.114s/rec
|
| 237 |
+
ep1 step 2330/3144 loss 0.282 kl 0.000 anchor 0.000 0.114s/rec
|
| 238 |
+
ep1 step 2340/3144 loss 0.381 kl 0.000 anchor 0.000 0.114s/rec
|
| 239 |
+
ep1 step 2350/3144 loss 0.323 kl 0.000 anchor 0.000 0.114s/rec
|
| 240 |
+
ep1 step 2360/3144 loss 0.363 kl 0.000 anchor 0.000 0.114s/rec
|
| 241 |
+
ep1 step 2370/3144 loss 0.371 kl 0.000 anchor 0.000 0.114s/rec
|
| 242 |
+
ep1 step 2380/3144 loss 0.349 kl 0.000 anchor 0.000 0.114s/rec
|
| 243 |
+
ep1 step 2390/3144 loss 0.160 kl 0.000 anchor 0.000 0.114s/rec
|
| 244 |
+
ep1 step 2400/3144 loss 0.196 kl 0.000 anchor 0.000 0.114s/rec
|
| 245 |
+
ep1 step 2410/3144 loss 0.165 kl 0.000 anchor 0.000 0.114s/rec
|
| 246 |
+
ep1 step 2420/3144 loss 0.222 kl 0.000 anchor 0.000 0.114s/rec
|
| 247 |
+
ep1 step 2430/3144 loss 0.236 kl 0.000 anchor 0.000 0.114s/rec
|
| 248 |
+
ep1 step 2440/3144 loss 0.213 kl 0.000 anchor 0.000 0.114s/rec
|
| 249 |
+
ep1 step 2450/3144 loss 0.139 kl 0.000 anchor 0.000 0.114s/rec
|
| 250 |
+
ep1 step 2460/3144 loss 0.182 kl 0.000 anchor 0.000 0.113s/rec
|
| 251 |
+
ep1 step 2470/3144 loss 0.379 kl 0.000 anchor 0.000 0.114s/rec
|
| 252 |
+
ep1 step 2480/3144 loss 0.248 kl 0.000 anchor 0.000 0.113s/rec
|
| 253 |
+
ep1 step 2490/3144 loss 0.318 kl 0.000 anchor 0.000 0.113s/rec
|
| 254 |
+
ep1 step 2500/3144 loss 0.238 kl 0.000 anchor 0.000 0.113s/rec
|
| 255 |
+
ep1 step 2510/3144 loss 0.322 kl 0.000 anchor 0.000 0.113s/rec
|
| 256 |
+
ep1 step 2520/3144 loss 0.466 kl 0.000 anchor 0.000 0.113s/rec
|
| 257 |
+
ep1 step 2530/3144 loss 0.267 kl 0.000 anchor 0.000 0.113s/rec
|
| 258 |
+
ep1 step 2540/3144 loss 0.302 kl 0.000 anchor 0.000 0.113s/rec
|
| 259 |
+
ep1 step 2550/3144 loss 0.260 kl 0.000 anchor 0.000 0.113s/rec
|
| 260 |
+
ep1 step 2560/3144 loss 0.249 kl 0.000 anchor 0.000 0.113s/rec
|
| 261 |
+
ep1 step 2570/3144 loss 0.514 kl 0.000 anchor 0.000 0.113s/rec
|
| 262 |
+
ep1 step 2580/3144 loss 0.146 kl 0.000 anchor 0.000 0.113s/rec
|
| 263 |
+
ep1 step 2590/3144 loss 0.314 kl 0.000 anchor 0.000 0.113s/rec
|
| 264 |
+
ep1 step 2600/3144 loss 0.306 kl 0.000 anchor 0.000 0.113s/rec
|
| 265 |
+
ep1 step 2610/3144 loss 0.537 kl 0.000 anchor 0.000 0.113s/rec
|
| 266 |
+
ep1 step 2620/3144 loss 0.237 kl 0.000 anchor 0.000 0.113s/rec
|
| 267 |
+
ep1 step 2630/3144 loss 0.136 kl 0.000 anchor 0.000 0.113s/rec
|
| 268 |
+
ep1 step 2640/3144 loss 0.270 kl 0.000 anchor 0.000 0.113s/rec
|
| 269 |
+
ep1 step 2650/3144 loss 0.252 kl 0.000 anchor 0.000 0.113s/rec
|
| 270 |
+
ep1 step 2660/3144 loss 0.254 kl 0.000 anchor 0.000 0.113s/rec
|
| 271 |
+
ep1 step 2670/3144 loss 0.270 kl 0.000 anchor 0.000 0.113s/rec
|
| 272 |
+
ep1 step 2680/3144 loss 0.218 kl 0.000 anchor 0.000 0.113s/rec
|
| 273 |
+
ep1 step 2690/3144 loss 0.382 kl 0.000 anchor 0.000 0.113s/rec
|
| 274 |
+
ep1 step 2700/3144 loss 0.362 kl 0.000 anchor 0.000 0.113s/rec
|
| 275 |
+
ep1 step 2710/3144 loss 0.380 kl 0.000 anchor 0.000 0.113s/rec
|
| 276 |
+
ep1 step 2720/3144 loss 0.426 kl 0.000 anchor 0.000 0.113s/rec
|
| 277 |
+
ep1 step 2730/3144 loss 0.120 kl 0.000 anchor 0.000 0.113s/rec
|
| 278 |
+
ep1 step 2740/3144 loss 0.417 kl 0.000 anchor 0.000 0.113s/rec
|
| 279 |
+
ep1 step 2750/3144 loss 0.183 kl 0.000 anchor 0.000 0.113s/rec
|
| 280 |
+
ep1 step 2760/3144 loss 0.393 kl 0.000 anchor 0.000 0.112s/rec
|
| 281 |
+
ep1 step 2770/3144 loss 0.180 kl 0.000 anchor 0.000 0.112s/rec
|
| 282 |
+
ep1 step 2780/3144 loss 0.128 kl 0.000 anchor 0.000 0.112s/rec
|
| 283 |
+
ep1 step 2790/3144 loss 0.154 kl 0.000 anchor 0.000 0.112s/rec
|
| 284 |
+
ep1 step 2800/3144 loss 0.139 kl 0.000 anchor 0.000 0.112s/rec
|
| 285 |
+
ep1 step 2810/3144 loss 0.184 kl 0.000 anchor 0.000 0.112s/rec
|
| 286 |
+
ep1 step 2820/3144 loss 0.167 kl 0.000 anchor 0.000 0.112s/rec
|
| 287 |
+
ep1 step 2830/3144 loss 0.373 kl 0.000 anchor 0.000 0.112s/rec
|
| 288 |
+
ep1 step 2840/3144 loss 0.275 kl 0.000 anchor 0.000 0.112s/rec
|
| 289 |
+
ep1 step 2850/3144 loss 0.268 kl 0.000 anchor 0.000 0.112s/rec
|
| 290 |
+
ep1 step 2860/3144 loss 0.253 kl 0.000 anchor 0.000 0.112s/rec
|
| 291 |
+
ep1 step 2870/3144 loss 0.446 kl 0.000 anchor 0.000 0.112s/rec
|
| 292 |
+
ep1 step 2880/3144 loss 0.449 kl 0.000 anchor 0.000 0.112s/rec
|
| 293 |
+
ep1 step 2890/3144 loss 0.495 kl 0.000 anchor 0.000 0.112s/rec
|
| 294 |
+
ep1 step 2900/3144 loss 0.430 kl 0.000 anchor 0.000 0.112s/rec
|
| 295 |
+
ep1 step 2910/3144 loss 0.123 kl 0.000 anchor 0.000 0.112s/rec
|
| 296 |
+
ep1 step 2920/3144 loss 0.113 kl 0.000 anchor 0.000 0.112s/rec
|
| 297 |
+
ep1 step 2930/3144 loss 0.155 kl 0.000 anchor 0.000 0.112s/rec
|
| 298 |
+
ep1 step 2940/3144 loss 0.340 kl 0.000 anchor 0.000 0.112s/rec
|
| 299 |
+
ep1 step 2950/3144 loss 0.274 kl 0.000 anchor 0.000 0.112s/rec
|
| 300 |
+
ep1 step 2960/3144 loss 0.251 kl 0.000 anchor 0.000 0.112s/rec
|
| 301 |
+
ep1 step 2970/3144 loss 0.248 kl 0.000 anchor 0.000 0.112s/rec
|
| 302 |
+
ep1 step 2980/3144 loss 0.125 kl 0.000 anchor 0.000 0.112s/rec
|
| 303 |
+
ep1 step 2990/3144 loss 0.370 kl 0.000 anchor 0.000 0.112s/rec
|
| 304 |
+
ep1 step 3000/3144 loss 0.432 kl 0.000 anchor 0.000 0.112s/rec
|
| 305 |
+
ep1 step 3010/3144 loss 0.305 kl 0.000 anchor 0.000 0.112s/rec
|
| 306 |
+
ep1 step 3020/3144 loss 0.339 kl 0.000 anchor 0.000 0.112s/rec
|
| 307 |
+
ep1 step 3030/3144 loss 0.333 kl 0.000 anchor 0.000 0.112s/rec
|
| 308 |
+
ep1 step 3040/3144 loss 0.141 kl 0.000 anchor 0.000 0.112s/rec
|
| 309 |
+
ep1 step 3050/3144 loss 0.176 kl 0.000 anchor 0.000 0.112s/rec
|
| 310 |
+
ep1 step 3060/3144 loss 0.263 kl 0.000 anchor 0.000 0.112s/rec
|
| 311 |
+
ep1 step 3070/3144 loss 0.160 kl 0.000 anchor 0.000 0.112s/rec
|
| 312 |
+
ep1 step 3080/3144 loss 0.332 kl 0.000 anchor 0.000 0.112s/rec
|
| 313 |
+
ep1 step 3090/3144 loss 0.209 kl 0.000 anchor 0.000 0.112s/rec
|
| 314 |
+
ep1 step 3100/3144 loss 0.247 kl 0.000 anchor 0.000 0.112s/rec
|
| 315 |
+
ep1 step 3110/3144 loss 0.356 kl 0.000 anchor 0.000 0.112s/rec
|
| 316 |
+
ep1 step 3120/3144 loss 0.311 kl 0.000 anchor 0.000 0.112s/rec
|
| 317 |
+
ep1 step 3130/3144 loss 0.222 kl 0.000 anchor 0.000 0.111s/rec
|
| 318 |
+
ep1 step 3140/3144 loss 0.187 kl 0.000 anchor 0.000 0.111s/rec
|
| 319 |
+
saved /runs/q35-4b-s23/00-trial-0/checkpoint
|
training_config.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"args": {
|
| 3 |
-
"base": "Qwen/Qwen3-4B-Base",
|
| 4 |
"n_per_source": 1000,
|
| 5 |
"epochs": 2,
|
| 6 |
"lr": 5e-05,
|
|
@@ -22,7 +22,7 @@
|
|
| 22 |
"special_embeddings": 0,
|
| 23 |
"head_dim": 256,
|
| 24 |
"lora_targets": "all",
|
| 25 |
-
"base_revision": "",
|
| 26 |
"p_none": 0.1,
|
| 27 |
"p_none_distract": 0.12,
|
| 28 |
"p_distract": 0.15,
|
|
@@ -32,11 +32,11 @@
|
|
| 32 |
"anchor": "",
|
| 33 |
"anchor_w": 0.0,
|
| 34 |
"anchor_sources": "",
|
| 35 |
-
"out": "/runs/
|
| 36 |
-
"seed":
|
| 37 |
},
|
| 38 |
"suite_sha256": "a8f50e481b7d90b97da049e0ff6a01cee2f1ed204aed61a8265af0edbb5514d2",
|
| 39 |
-
"base_revision": "
|
| 40 |
"ordinal_objective": "ranked_probability_score",
|
| 41 |
"holdout": []
|
| 42 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"args": {
|
| 3 |
+
"base": "Qwen/Qwen3.5-4B-Base",
|
| 4 |
"n_per_source": 1000,
|
| 5 |
"epochs": 2,
|
| 6 |
"lr": 5e-05,
|
|
|
|
| 22 |
"special_embeddings": 0,
|
| 23 |
"head_dim": 256,
|
| 24 |
"lora_targets": "all",
|
| 25 |
+
"base_revision": "1001bb4d826a52d1f399e183466143f4da7b741b",
|
| 26 |
"p_none": 0.1,
|
| 27 |
"p_none_distract": 0.12,
|
| 28 |
"p_distract": 0.15,
|
|
|
|
| 32 |
"anchor": "",
|
| 33 |
"anchor_w": 0.0,
|
| 34 |
"anchor_sources": "",
|
| 35 |
+
"out": "/runs/q35-4b-s23/00-trial-0/checkpoint",
|
| 36 |
+
"seed": 2
|
| 37 |
},
|
| 38 |
"suite_sha256": "a8f50e481b7d90b97da049e0ff6a01cee2f1ed204aed61a8265af0edbb5514d2",
|
| 39 |
+
"base_revision": "1001bb4d826a52d1f399e183466143f4da7b741b",
|
| 40 |
"ordinal_objective": "ranked_probability_score",
|
| 41 |
"holdout": []
|
| 42 |
}
|
training_metrics.json
CHANGED
|
@@ -1,14 +1,14 @@
|
|
| 1 |
{
|
| 2 |
-
"wall_seconds":
|
| 3 |
-
"records_seen":
|
| 4 |
"requested_records": 25152,
|
| 5 |
"truncated_records": 0,
|
| 6 |
"rejected_records": 0,
|
| 7 |
"optimizer_steps": 3144,
|
| 8 |
-
"forward_tokens":
|
| 9 |
-
"peak_device_bytes":
|
| 10 |
"device": "cuda",
|
| 11 |
"dtype": "bf16",
|
| 12 |
"batch": 4,
|
| 13 |
-
"peak_rss_bytes":
|
| 14 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"wall_seconds": 3384.1752576828003,
|
| 3 |
+
"records_seen": 30370,
|
| 4 |
"requested_records": 25152,
|
| 5 |
"truncated_records": 0,
|
| 6 |
"rejected_records": 0,
|
| 7 |
"optimizer_steps": 3144,
|
| 8 |
+
"forward_tokens": 5832366,
|
| 9 |
+
"peak_device_bytes": 24631309312,
|
| 10 |
"device": "cuda",
|
| 11 |
"dtype": "bf16",
|
| 12 |
"batch": 4,
|
| 13 |
+
"peak_rss_bytes": 31542767616
|
| 14 |
}
|