diff options
Diffstat (limited to 'hyperstack-vm2.toml')
| -rw-r--r-- | hyperstack-vm2.toml | 1 |
1 files changed, 1 insertions, 0 deletions
diff --git a/hyperstack-vm2.toml b/hyperstack-vm2.toml index d3c0a17..32e3a99 100644 --- a/hyperstack-vm2.toml +++ b/hyperstack-vm2.toml @@ -116,6 +116,7 @@ max_model_len = 131072 gpu_memory_utilization = 0.92 tensor_parallel_size = 1 tool_call_parser = "" +extra_vllm_args = ["--reasoning-parser", "openai_gptoss"] # Qwen2.5-Coder-32B-Instruct AWQ — best-in-class open coding model at 32B, ~18 GB on A100. [vllm.presets.qwen25-coder-32b] |