pivot judge audit: selector/20261009-selector/qwen3

selector/20261009-selector/qwen3/code.sha256

@@ -0,0 +1 @@
+5de108b619ec0e2535dea6a02df9ae51b4aa98030fe26b3b8b184e58827470be  /dev/volume/outputs/pivot_judge/selector/20261009-selector/qwen3/code.tar.gz

selector/20261009-selector/qwen3/code.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5de108b619ec0e2535dea6a02df9ae51b4aa98030fe26b3b8b184e58827470be
+size 87289

selector/20261009-selector/qwen3/env.json

@@ -0,0 +1,13 @@
+{
+  "host": "pivotopd-selector-4dwgs-vfjjc",
+  "python": "3.12.15",
+  "samples": 5,
+  "variants": "terse,fenced",
+  "max_model_len": 16384,
+  "source_run": "results/20261006-base-r1/qwen3",
+  "vllm": "0.11.0",
+  "torch": "2.8.0+cu128",
+  "transformers": "4.57.1",
+  "huggingface_hub": "0.36.2",
+  "gpu": "NVIDIA A100-SXM4-80GB, 81920 MiB"
+}

selector/20261009-selector/qwen3/selector.log

@@ -0,0 +1,45 @@
+INFO 10-08 23:39:06 [__init__.py:216] Automatically detected platform cuda.
+[pivot_judge 23:39:11] [selector] 320 requests (320 to run) over 32 successful parents; max prompt 2624 tokens, budget 8192
+INFO 10-08 23:39:11 [utils.py:233] non-default args: {'dtype': 'bfloat16', 'seed': 0, 'max_model_len': 16384, 'enable_prefix_caching': True, 'gpu_memory_utilization': 0.85, 'disable_log_stats': True, 'model': 'Qwen/Qwen3-8B'}
+INFO 10-08 23:39:25 [model.py:547] Resolved architecture: Qwen3ForCausalLM
+`torch_dtype` is deprecated! Use `dtype` instead!
+INFO 10-08 23:39:25 [model.py:1510] Using max model len 16384
+INFO 10-08 23:39:25 [scheduler.py:205] Chunked prefill is enabled with max_num_batched_tokens=8192.
+(EngineCore_DP0 pid=8786) INFO 10-08 23:39:26 [core.py:644] Waiting for init message from front-end.
+(EngineCore_DP0 pid=8786) INFO 10-08 23:39:26 [core.py:77] Initializing a V1 LLM engine (v0.11.0) with config: model='Qwen/Qwen3-8B', speculative_config=None, tokenizer='Qwen/Qwen3-8B', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, tokenizer_revision=None, trust_remote_code=False, dtype=torch.bfloat16, max_seq_len=16384, download_dir=None, load_format=auto, tensor_parallel_size=1, pipeline_parallel_size=1, data_parallel_size=1, disable_custom_all_reduce=False, quantization=None, enforce_eager=False, kv_cache_dtype=auto, device_config=cuda, structured_outputs_config=StructuredOutputsConfig(backend='auto', disable_fallback=False, disable_any_whitespace=False, disable_additional_properties=False, reasoning_parser=''), observability_config=ObservabilityConfig(show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None), seed=0, served_model_name=Qwen/Qwen3-8B, enable_prefix_caching=True, chunked_prefill_enabled=True, pooler_config=None, compilation_config={"level":3,"debug_dump_path":"","cache_dir":"","backend":"","custom_ops":[],"splitting_ops":["vllm.unified_attention","vllm.unified_attention_with_output","vllm.mamba_mixer2","vllm.mamba_mixer","vllm.short_conv","vllm.linear_attention","vllm.plamo2_mamba_mixer","vllm.gdn_attention","vllm.sparse_attn_indexer"],"use_inductor":true,"compile_sizes":[],"inductor_compile_config":{"enable_auto_functionalized_v2":false},"inductor_passes":{},"cudagraph_mode":[2,1],"use_cudagraph":true,"cudagraph_num_of_warmups":1,"cudagraph_capture_sizes":[512,504,496,488,480,472,464,456,448,440,432,424,416,408,400,392,384,376,368,360,352,344,336,328,320,312,304,296,288,280,272,264,256,248,240,232,224,216,208,200,192,184,176,168,160,152,144,136,128,120,112,104,96,88,80,72,64,56,48,40,32,24,16,8,4,2,1],"cudagraph_copy_inputs":false,"full_cuda_graph":false,"use_inductor_graph_partition":false,"pass_config":{},"max_capture_size":512,"local_cache_dir":null}
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0
+(EngineCore_DP0 pid=8786) INFO 10-08 23:39:27 [parallel_state.py:1208] rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, TP rank 0, EP rank 0
+(EngineCore_DP0 pid=8786) WARNING 10-08 23:39:28 [topk_topp_sampler.py:66] FlashInfer is not available. Falling back to the PyTorch-native implementation of top-p & top-k sampling. For the best performance, please install FlashInfer.
+(EngineCore_DP0 pid=8786) INFO 10-08 23:39:29 [gpu_model_runner.py:2602] Starting to load model Qwen/Qwen3-8B...
+(EngineCore_DP0 pid=8786) INFO 10-08 23:39:29 [gpu_model_runner.py:2634] Loading model from scratch...
+(EngineCore_DP0 pid=8786) INFO 10-08 23:39:30 [cuda.py:366] Using Flash Attention backend on V1 engine.
+(EngineCore_DP0 pid=8786) INFO 10-08 23:39:30 [weight_utils.py:392] Using model weights format ['*.safetensors']
+(EngineCore_DP0 pid=8786) 
+(EngineCore_DP0 pid=8786) 
+(EngineCore_DP0 pid=8786) 
+(EngineCore_DP0 pid=8786) 
+(EngineCore_DP0 pid=8786) 
+(EngineCore_DP0 pid=8786) 
+(EngineCore_DP0 pid=8786) 
+(EngineCore_DP0 pid=8786) 
+(EngineCore_DP0 pid=8786) INFO 10-08 23:40:26 [default_loader.py:267] Loading weights took 55.64 seconds
+(EngineCore_DP0 pid=8786) INFO 10-08 23:40:26 [gpu_model_runner.py:2653] Model loading took 15.2683 GiB and 56.506072 seconds
+(EngineCore_DP0 pid=8786) INFO 10-08 23:40:38 [backends.py:548] Using cache directory: /root/.cache/vllm/torch_compile_cache/af0af28466/rank_0_0/backbone for vLLM's torch.compile
+(EngineCore_DP0 pid=8786) INFO 10-08 23:40:38 [backends.py:559] Dynamo bytecode transform time: 10.81 s
+(EngineCore_DP0 pid=8786) INFO 10-08 23:40:44 [backends.py:197] Cache the graph for dynamic shape for later use
+(EngineCore_DP0 pid=8786) INFO 10-08 23:41:20 [backends.py:218] Compiling a graph for dynamic shape takes 41.02 s
+(EngineCore_DP0 pid=8786) INFO 10-08 23:41:36 [monitor.py:34] torch.compile takes 51.83 s in total
+(EngineCore_DP0 pid=8786) INFO 10-08 23:41:38 [gpu_worker.py:298] Available KV cache memory: 50.60 GiB
+(EngineCore_DP0 pid=8786) INFO 10-08 23:41:38 [kv_cache_utils.py:1087] GPU KV cache size: 368,480 tokens
+(EngineCore_DP0 pid=8786) INFO 10-08 23:41:38 [kv_cache_utils.py:1091] Maximum concurrency for 16,384 tokens per request: 22.49x
+(EngineCore_DP0 pid=8786) 
+(EngineCore_DP0 pid=8786) 
+(EngineCore_DP0 pid=8786) INFO 10-08 23:41:47 [gpu_model_runner.py:3480] Graph capturing finished in 9 secs, took 0.85 GiB
+(EngineCore_DP0 pid=8786) INFO 10-08 23:41:47 [core.py:210] init engine (profile, create kv cache, warmup model) took 80.65 seconds
+INFO 10-08 23:41:49 [llm.py:306] Supported_tasks: ['generate']
+[pivot_judge 23:54:21] [selector] outcomes {'terse': {'picked': 156, 'no_valid_id': 2, 'response_truncated': 2}, 'fenced': {'response_truncated': 6, 'picked': 154}} in 911.8 s

selector/20261009-selector/qwen3/selector_meta.json

@@ -0,0 +1,34 @@
+{
+  "model": "Qwen/Qwen3-8B",
+  "enable_thinking": "true",
+  "settings": {
+    "obs_char_cap": 500,
+    "obs_fit": true,
+    "max_prompt_tokens": 4096,
+    "temperature": 1.0,
+    "top_p": 1.0,
+    "top_k": -1,
+    "max_response_tokens": 8192
+  },
+  "max_model_len": 16384,
+  "max_tokens": 8192,
+  "samples": 5,
+  "variants": [
+    "terse",
+    "fenced"
+  ],
+  "seed": 0,
+  "selector_source_sha256": "0655eefe086a3fa0c1a33d1e581b3c265f95699d62f53f04d06976642fac24a5",
+  "outcomes": {
+    "terse": {
+      "picked": 156,
+      "no_valid_id": 2,
+      "response_truncated": 2
+    },
+    "fenced": {
+      "response_truncated": 6,
+      "picked": 154
+    }
+  },
+  "seconds": 911.8
+}