| { |
| "scope": "orcarouter/Qwen3.8-Flash-Next-Uncensored-NVFP4 only", |
| "target_revision": "3a3b63161c0745390e5270179af42e46efc70799", |
| "mtp_source": "RadixArk/Qwen3.8-Flash-Next-NVFP4", |
| "mtp_source_revision": "7b719225242aacd3dbd3f9407468c2ee9a9d2594", |
| "runtime": "vLLM 0.1.dev20073+g8e685d198", |
| "context_window": 32768, |
| "concurrency": 1, |
| "max_num_seqs": 1, |
| "gpu_memory_utilization": 0.8, |
| "sampling": { |
| "thinking": true, |
| "reasoning_effort": "medium", |
| "temperature": 1.0, |
| "top_p": 0.95, |
| "top_k": 20, |
| "min_p": 0.0, |
| "presence_penalty": 0.0, |
| "repetition_penalty": 1.0 |
| }, |
| "measurement": "single-stream median of three fixed 256-token streamed samples after a 128-token warm-up", |
| "timing_definitions": { |
| "end_to_end_completion_tokens_per_second": "all completion tokens, including hidden reasoning, divided by total request wall time", |
| "time_to_first_visible_content_seconds": "request start through the first non-empty visible content delta", |
| "visible_content_tokens_per_second": "visible content tokens only, measured from the first through last visible content delta" |
| }, |
| "selected_depth": 2, |
| "selected_depth_end_to_end_gain_vs_mtp0": 0.6222, |
| "mtp_overlay": { |
| "dtype": "BF16", |
| "loaded_tensors": 31, |
| "expected_tensors": 31, |
| "compact_file_gib": 4.86 |
| }, |
| "results": [ |
| {"depth": 0, "median_end_to_end_completion_tokens_per_second": 27.2645, "median_time_to_first_visible_content_seconds": 1.1963, "acceptance": null}, |
| {"depth": 1, "median_end_to_end_completion_tokens_per_second": 33.1749, "median_time_to_first_visible_content_seconds": 1.20583, "accepted_tokens": 305, "draft_tokens": 463, "acceptance": 0.6587}, |
| {"depth": 2, "median_end_to_end_completion_tokens_per_second": 44.2286, "median_visible_content_tokens_per_second": 46.1523, "median_time_to_first_visible_content_seconds": 1.32445, "accepted_tokens": 532, "draft_tokens": 722, "acceptance": 0.7368}, |
| {"depth": 3, "median_end_to_end_completion_tokens_per_second": 40.9544, "median_time_to_first_visible_content_seconds": 1.62869, "accepted_tokens": 563, "draft_tokens": 993, "acceptance": 0.5670}, |
| {"depth": 4, "median_end_to_end_completion_tokens_per_second": 41.8389, "median_time_to_first_visible_content_seconds": 1.12255, "accepted_tokens": 611, "draft_tokens": 1148, "acceptance": 0.5322}, |
| {"depth": 5, "median_end_to_end_completion_tokens_per_second": 35.4522, "median_time_to_first_visible_content_seconds": 1.24926, "accepted_tokens": 454, "draft_tokens": 1560, "acceptance": 0.2910}, |
| {"depth": 6, "median_end_to_end_completion_tokens_per_second": 35.7146, "median_time_to_first_visible_content_seconds": 1.42869, "accepted_tokens": 516, "draft_tokens": 1230, "acceptance": 0.4195}, |
| {"depth": 8, "median_end_to_end_completion_tokens_per_second": 30.4295, "median_time_to_first_visible_content_seconds": 1.75837, "accepted_tokens": 487, "draft_tokens": 1600, "acceptance": 0.3044}, |
| {"depth": 10, "median_end_to_end_completion_tokens_per_second": 28.3075, "median_time_to_first_visible_content_seconds": 1.64792, "accepted_tokens": 642, "draft_tokens": 2150, "acceptance": 0.2986} |
| ], |
| "native_context_validation": { |
| "max_model_len": 262144, |
| "max_num_seqs": 8, |
| "gpu_memory_utilization": 0.8, |
| "model_memory_gib": 76.21, |
| "available_kv_gib": 17.37, |
| "kv_capacity_tokens": 627960, |
| "maximum_concurrency_at_native_context": 2.4, |
| "prompt_tokens": 240051, |
| "exact_retrieval": true, |
| "ttft_seconds": 116.8036, |
| "prefill_tokens_per_second": 2055.17 |
| } |
| } |
|
|
