Issues

• [Bug] Qwen3.5/3.6 (GDN): prompts longer than gdn_chunk_size produce broken output on Cloud AI 100 when prefill_seq_len > gdn_chunk_size

  qualcomm/efficient-transformers#1446 · Filter by author ronaldescalona opened on Oct 21, 2026

• [Bug] ONNX export of decoder-based embedding models (Qwen3-Embedding) fails with transformers 5.5.4: IndexError in masking_utils.sdpa_mask

  qualcomm/efficient-transformers#1445 · Filter by author ronaldescalona opened on Oct 21, 2026

• Layerwise compilation fails for MoE model with continuous batching: batch_index missing from split subgraph bounds

  Stale
  qualcomm/efficient-transformers#1122 · Filter by author weimin023 opened on Jun 25, 2026
  3

• Need GLM-4.7-Flash Support

  Stale
  Feature

  qualcomm/efficient-transformers#850 · Filter by author azajac-qcom opened on Mar 13, 2026
  3

• Use keyword argument for export_dir in ONNX export calls to avoid positional misbinding.

  Stale
  qualcomm/efficient-transformers#798 · Filter by author jatin5jan opened on Feb 17, 2026
  1

• Fail to compile Gemma3n model

  Stale
  qualcomm/efficient-transformers#518 · Filter by author lehuudoUIT opened on Jul 16, 2025
  3