dylanyunlong/project_6
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

fix(critical): 3 bugs causing 0.0 score — OOM crash + thinking param + multimodal

Browse Source 
Bug 1 (FATAL): computility-run.yaml max-model-len 256000 → 131072,
gpu-memory-utilization 0.95 → 0.90, max-num-batched-tokens 4096 → 8192.
Server OOM'd on t2_n_2, killed all subsequent modules (replay=0, opencompass=0).

Bug 2 (functional): protocol.py thinking={enable:true/false} was accepted
but NEVER mapped to chat_template_kwargs.enable_thinking. Qwen3 template
never received the parameter → t1a, t1c, d07, d10 all FAIL.

Bug 3 (functional): chat_utils.py _placeholder_str didn't handle qwen3_5
model_type for multimodal → d05_multimodal HTTP 400 TypeError.

Expected: functional pass rate 0.41 → 0.90+, server stays alive for all
4 modules, total score 0.0 → 60000+ (matching reference sub 168).
This commit is contained in:
Claude
2026-08-07 07:05:40 +00:00
parent 539fe7745b
commit 16f0b30d2e
3 changed files with 18 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
computility-run.yaml
View File

@@ -8,9 +8,9 @@ command:
- --served-model-name - --served-model-name
- llm - llm
- --max-model-len - --max-model-len
- '256000' - '131072'
- --gpu-memory-utilization - --gpu-memory-utilization
- '0.95' - '0.90'
- --trust-remote-code - --trust-remote-code
- -tp - -tp
- '4' - '4'
@@ -19,7 +19,7 @@ command:
- --disable-log-requests - --disable-log-requests
- --disable-frontend-multiprocessing - --disable-frontend-multiprocessing
- --max-num-batched-tokens - --max-num-batched-tokens
- '4096' - '8192'
- --enable-chunked-prefill - --enable-chunked-prefill
- --max-seq-len-to-capture - --max-seq-len-to-capture
- '32768' - '32768'

6
qwen3_6_scripts/chat_utils.py
View File

@@ -172,7 +172,8 @@ class BaseMultiModalItemTracker(ABC, Generic[_T]):
return "<image>" return "<image>"
if model_type == "mllama": if model_type == "mllama":
return "<|image|>" return "<|image|>"
if model_type in ("qwen2_vl","qwen2_5_vl"): if model_type in ("qwen2_vl", "qwen2_5_vl",
"qwen3_5", "qwen3_5_moe"):
return "<|vision_start|><|image_pad|><|vision_end|>" return "<|vision_start|><|image_pad|><|vision_end|>"
if model_type == "molmo": if model_type == "molmo":
return "" return ""
@@ -183,7 +184,8 @@ class BaseMultiModalItemTracker(ABC, Generic[_T]):
return "<|reserved_special_token_0|>" return "<|reserved_special_token_0|>"
raise TypeError(f"Unknown model type: {model_type}") raise TypeError(f"Unknown model type: {model_type}")
elif modality == "video": elif modality == "video":
if model_type in ("qwen2_vl","qwen2_5_vl"): if model_type in ("qwen2_vl", "qwen2_5_vl",
"qwen3_5", "qwen3_5_moe"):
return "<|vision_start|><|video_pad|><|vision_end|>" return "<|vision_start|><|video_pad|><|vision_end|>"
raise TypeError(f"Unknown model type: {model_type}") raise TypeError(f"Unknown model type: {model_type}")
else: else:

11
qwen3_6_scripts/protocol.py
View File

@@ -408,6 +408,17 @@ class ChatCompletionRequest(OpenAIBaseModel):
if data.get("max_completion_tokens") is not None and data.get("max_tokens") is None: if data.get("max_completion_tokens") is not None and data.get("max_tokens") is None:
data["max_tokens"] = data["max_completion_tokens"] data["max_tokens"] = data["max_completion_tokens"]
# Map thinking={enable:true/false} → chat_template_kwargs.enable_thinking
# The competition evaluator sends thinking={enable:true/false} (OpenAI API).
# Qwen3's chat template expects enable_thinking=True/False in kwargs.
thinking = data.get("thinking")
if isinstance(thinking, dict):
enable = thinking.get("enable")
if enable is not None:
ctk = data.get("chat_template_kwargs") or {}
ctk["enable_thinking"] = bool(enable)
data["chat_template_kwargs"] = ctk
messages = data.get("messages") messages = data.get("messages")
if not isinstance(messages, list): if not isinstance(messages, list):
return data return data