| { |
| "dataset": { |
| "examples": 1319, |
| "harness": "sgl-eval", |
| "name": "gsm8k", |
| "revision": "6690895609dcbc5df1e7b00dd57c9502b868ec4d" |
| }, |
| "gates": { |
| "empty_generations": 0, |
| "maximum_request_errors": 0, |
| "maximum_truncations": 0, |
| "minimum_accuracy": 0.965, |
| "minimum_mtp_acceptance_length": 1.0, |
| "stop_rate": 1.0, |
| "target_oom_kills": 0 |
| }, |
| "runtime": { |
| "cuda_graph_max_batch_size": 64, |
| "kv_cache_dtype": "fp8_e4m3", |
| "mamba_radix_cache_strategy": "extra_buffer", |
| "memory_fraction_static": 0.75, |
| "modelopt_loader": "modelopt_fp4", |
| "resolved_backends": "automatic", |
| "sglang_revision": "29481685462732237d80d86076d6563e1f658102", |
| "speculative_algorithm": "NEXTN", |
| "speculative_draft_tokens": 4, |
| "speculative_num_steps": 3, |
| "speculative_topk": 1, |
| "tensor_parallel_size": 4 |
| }, |
| "sampling": { |
| "min_p": 0.0, |
| "presence_penalty": 0.0, |
| "repetition_penalty": 1.0, |
| "temperature": 1.0, |
| "thinking": true, |
| "top_k": 20, |
| "top_p": 0.95 |
| }, |
| "smoke": { |
| "exact_answer": "5", |
| "finite_logprobs": true, |
| "finish_reason": "stop" |
| } |
| } |
|
|
