Segment 15
The assistant fixed the `KeyError: qweight_type` by force-dequantizing tensors with `quant_config=None` (e.g., `weights_proj`, MoE gate) and skipping unknown parameter names in `load_weights`. After deploying these patches, the GLM-5 GGUF model loaded fully and the vLLM server began serving requests. However, the generated output was incoherent with flat log-prob distributions. The assistant systematically investigated possible causes: confirmed the GGUF dequantization kernel works on SM120, verified weight name mapping correctness, and confirmed FlashAttention availability. The investigation narrowed to the `kv_b_proj` weight loading: the weight is reassembled as a full `[28672, 512]` tensor but the `ColumnParallelLinear` expects a TP-sharded `[3584, 512]` parameter, yet no assertion error occurred, suggesting the parameter might be `UninitializedParameter` or the loader handles the mismatch differently. The assistant was preparing to further investigate this sharding discrepancy.
From Crash to Coherence: Debugging the GLM-5 GGUF Deployment on 8× Blackwell GPUs