{ "context": { "cap_tokens": null, "max_model_len": null, "targets": [] }, "evaluation_protocol": { "budget_semantics": "visible_answer_tokens plus reasoning_headroom_tokens are sent as one max_tokens cap; the endpoint does not hard-partition the channels", "minimum_comparable_repetitions": 3, "minimum_pass_rate": 1.0, "reasoning_headroom_tokens": 8192, "records_finish_reason": true, "records_full_visible_answer": true, "records_reasoning_channel_metadata": true, "repetitions": 1, "version": 3, "visible_answer_tokens": 1024 }, "failures": [ { "error": "answer", "eval_id": "mmlu-pro-30-computer-science", "failure_classes": [ "reasoning_budget_exhausted" ], "pass_rate": 0.0, "suite": "mmlu-pro-computer-science-budget-retry" } ], "identity": { "base_url": "http://127.0.0.1:8081/v1", "candidate_id": "signal38-nospec", "config_id": "q6k-nospec-64k-budget-retry", "engine": "llama.cpp", "gpu": "rtx5090-primary", "model": "signal38-27b-q6k-nospec-64k", "started_at": "2026-09-12T19:33:02Z" }, "intelligence": { "checks": [], "status": "not_run" }, "run_id": "fast-bakeoff-20260912T193302Z", "schema": "anvil-serving.fast-tier-bakeoff/v1", "score_inputs": { "e2e_p50_ms": null, "intelligence_pass_rate": null, "operational_fit_notes": [ "endpoint was already loaded; benchmark did not start or stop serves" ], "session_recall_passed": false, "thinking_mode": "enabled", "tool_call_passed": false, "ttft_p50_ms": null, "usable_context_tokens": null, "voice_latency_ms": null }, "selection": { "context_targets": [ 32768 ], "endpoint_already_loaded": true, "requests_per_context": 1, "suites": [] }, "session": { "checks": [], "status": "not_run" }, "source_recipe": { "ref": "configs/qwen38-efficient-rtx5090-nospec-controls.toml", "serve_command": null }, "suites": { "mmlu-pro-computer-science-budget-retry": { "checks": [ { "attempt_count": 1, "attempts": [ { "attempt": 1, "budget": { "legacy_max_tokens_as_visible": false, "max_completion_tokens": 9216, "reasoning_headroom_tokens": 8192, "visible_answer_tokens": 1024 }, "content": "", "content_chars": 0, "content_excerpt": "", "error": "answer", "failure_class": "reasoning_budget_exhausted", "finish_reason": "length", "latency_ms": 167328.98829999613, "reasoning_chars": 32799, "reasoning_excerpt": "We need answer multiple choice. Need reason. Let's parse problem carefully.\n\nPipelined RISC machine registers R0-R7. Supports ADD Rs1, Rs2, Rd and MUL Rs1, Rs2, Rd. Operation normally takes one cycle;", "reasoning_field": "reasoning_content", "reasoning_tokens": null, "status": "failed", "text_checks": [ { "name": "answer", "passed": false } ], "usage": { "completion_tokens": 9216, "prompt_tokens": 220, "prompt_tokens_details": { "cached_tokens": 0 }, "total_tokens": 9436 } } ], "budget": { "legacy_max_tokens_as_visible": false, "max_completion_tokens": 9216, "reasoning_headroom_tokens": 8192, "visible_answer_tokens": 1024 }, "content": "", "content_excerpt": "", "error": "answer", "failure_class": "reasoning_budget_exhausted", "finish_reason": "length", "id": "mmlu-pro-30-computer-science", "latency_ms": 167328.98829999613, "pass_count": 0, "pass_rate": 0.0, "reasoning_chars": 32799, "reasoning_excerpt": "We need answer multiple choice. Need reason. Let's parse problem carefully.\n\nPipelined RISC machine registers R0-R7. Supports ADD Rs1, Rs2, Rd and MUL Rs1, Rs2, Rd. Operation normally takes one cycle;", "reasoning_field": "reasoning_content", "reasoning_tokens": null, "required_pass_rate": 1.0, "status": "failed", "text_checks": [ { "name": "answer", "passed": false } ], "usage": { "completion_tokens": 9216, "prompt_tokens": 220, "prompt_tokens_details": { "cached_tokens": 0 }, "total_tokens": 9436 }, "validator": "deterministic_text_checks" } ], "date": "2026-09-12", "evidence_use": "diagnostic", "source": "docs/findings/2026-09-12-qwen38-efficient-variants-rtx5090-evidence/mmlu-budget-retry.suite.json", "source_sha256": "c209f92737d52c39d78c875be3b30ac66528ec978d6295c5e57be63eabdb2d9f", "status": "failed", "validator_strength": "deterministic_marker", "work_class": "independent-multidomain-reasoning-sanity" } }, "thinking": { "chat_template_kwargs": { "enable_thinking": true }, "control_evidence": null, "control_evidence_sha256": null, "control_mechanism": "chat_template_kwargs", "control_requested": { "enable_thinking": true }, "control_status": "requested_unverified", "mode": "enabled", "reasoning_effort": null, "unsupported": false }, "timing": { "chat": { "completion_tokens_max": null, "completion_tokens_mean": null, "completion_tokens_mean_ci95_high": null, "completion_tokens_mean_ci95_low": null, "completion_tokens_min": null, "completion_tokens_p25": null, "completion_tokens_p50": null, "completion_tokens_p75": null, "completion_tokens_p90": null, "completion_tokens_p95": null, "completion_tokens_p99": null, "completion_tokens_samples": 0, "completion_tokens_stddev": null, "content_chunks": 0, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": null, "decode_tok_s_mean": null, "decode_tok_s_mean_ci95_high": null, "decode_tok_s_mean_ci95_low": null, "decode_tok_s_min": null, "decode_tok_s_p25": null, "decode_tok_s_p50": null, "decode_tok_s_p75": null, "decode_tok_s_p90": null, "decode_tok_s_p95": null, "decode_tok_s_p99": null, "decode_tok_s_samples": 0, "decode_tok_s_stddev": null, "e2e_max_ms": null, "e2e_mean_ci95_high_ms": null, "e2e_mean_ci95_low_ms": null, "e2e_mean_ms": null, "e2e_min_ms": null, "e2e_p25_ms": null, "e2e_p50_ms": null, "e2e_p75_ms": null, "e2e_p90_ms": null, "e2e_p95_ms": null, "e2e_p99_ms": null, "e2e_samples": 0, "e2e_stddev_ms": null, "effective_prefill_tok_s_max": null, "effective_prefill_tok_s_mean": null, "effective_prefill_tok_s_mean_ci95_high": null, "effective_prefill_tok_s_mean_ci95_low": null, "effective_prefill_tok_s_min": null, "effective_prefill_tok_s_p25": null, "effective_prefill_tok_s_p50": null, "effective_prefill_tok_s_p75": null, "effective_prefill_tok_s_p90": null, "effective_prefill_tok_s_p95": null, "effective_prefill_tok_s_p99": null, "effective_prefill_tok_s_samples": 0, "effective_prefill_tok_s_stddev": null, "generation_max_ms": null, "generation_mean_ci95_high_ms": null, "generation_mean_ci95_low_ms": null, "generation_mean_ms": null, "generation_min_ms": null, "generation_p25_ms": null, "generation_p50_ms": null, "generation_p75_ms": null, "generation_p90_ms": null, "generation_p95_ms": null, "generation_p99_ms": null, "generation_samples": 0, "generation_stddev_ms": null, "mean_inter_token_latency_ms_max": null, "mean_inter_token_latency_ms_mean": null, "mean_inter_token_latency_ms_mean_ci95_high": null, "mean_inter_token_latency_ms_mean_ci95_low": null, "mean_inter_token_latency_ms_min": null, "mean_inter_token_latency_ms_p25": null, "mean_inter_token_latency_ms_p50": null, "mean_inter_token_latency_ms_p75": null, "mean_inter_token_latency_ms_p90": null, "mean_inter_token_latency_ms_p95": null, "mean_inter_token_latency_ms_p99": null, "mean_inter_token_latency_ms_samples": 0, "mean_inter_token_latency_ms_stddev": null, "mean_time_per_output_token_max_ms": null, "mean_time_per_output_token_mean_ci95_high_ms": null, "mean_time_per_output_token_mean_ci95_low_ms": null, "mean_time_per_output_token_mean_ms": null, "mean_time_per_output_token_min_ms": null, "mean_time_per_output_token_p25_ms": null, "mean_time_per_output_token_p50_ms": null, "mean_time_per_output_token_p75_ms": null, "mean_time_per_output_token_p90_ms": null, "mean_time_per_output_token_p95_ms": null, "mean_time_per_output_token_p99_ms": null, "mean_time_per_output_token_samples": 0, "mean_time_per_output_token_stddev_ms": null, "output_tokens": 0, "prompt_token_samples": 0, "prompt_tokens": null, "prompt_tokens_max": null, "prompt_tokens_mean": null, "prompt_tokens_mean_ci95_high": null, "prompt_tokens_mean_ci95_low": null, "prompt_tokens_min": null, "prompt_tokens_p25": null, "prompt_tokens_p50": null, "prompt_tokens_p75": null, "prompt_tokens_p90": null, "prompt_tokens_p95": null, "prompt_tokens_p99": null, "prompt_tokens_samples": 0, "prompt_tokens_stddev": null, "reasoning_chunks": 0, "request_canary_passed": 0, "request_canary_samples": 0, "time_to_first_output_max_ms": null, "time_to_first_output_mean_ci95_high_ms": null, "time_to_first_output_mean_ci95_low_ms": null, "time_to_first_output_mean_ms": null, "time_to_first_output_min_ms": null, "time_to_first_output_p25_ms": null, "time_to_first_output_p50_ms": null, "time_to_first_output_p75_ms": null, "time_to_first_output_p90_ms": null, "time_to_first_output_p95_ms": null, "time_to_first_output_p99_ms": null, "time_to_first_output_samples": 0, "time_to_first_output_stddev_ms": null, "tpot_max_ms": null, "tpot_mean_ci95_high_ms": null, "tpot_mean_ci95_low_ms": null, "tpot_mean_ms": null, "tpot_min_ms": null, "tpot_p25_ms": null, "tpot_p50_ms": null, "tpot_p75_ms": null, "tpot_p90_ms": null, "tpot_p95_ms": null, "tpot_p99_ms": null, "tpot_samples": 0, "tpot_stddev_ms": null, "ttft_max_ms": null, "ttft_mean_ci95_high_ms": null, "ttft_mean_ci95_low_ms": null, "ttft_mean_ms": null, "ttft_min_ms": null, "ttft_p25_ms": null, "ttft_p50_ms": null, "ttft_p75_ms": null, "ttft_p90_ms": null, "ttft_p95_ms": null, "ttft_p99_ms": null, "ttft_samples": 0, "ttft_stddev_ms": null }, "wall_ms": 167350.33039998962 }, "tool": { "checks": [], "status": "not_run" }, "voice": { "status": "not_run", "stt_latency_ms": null, "total_turn_latency_ms": null, "tts_latency_ms": null } }