Repository navigation
docs: document Qwen3.8-Flash-Next as blocked (qwen4_exp architecture) - #52
Merged
Merged
Conversation
Verified via live onnxruntime-genai builder run (config_only) against Qwen/Qwen3.8-Flash-Next: the qwen4_exp architecture is not recognized by transformers 5.14.1 and has zero footprint in onnxruntime-genai's builder source (indexer_budget, hc_lowrank, ngram_vocab_size_base, qwen4_exp all return 0 GitHub code search results). Adds a new blocked-models.md entry following the Nemotron/Muse Glimmer template. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
… unmerged PR, not mainline
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Investigated adding Qwen3.8-Flash-Next (
unsloth/Qwen3.8-Flash-Next-GGUF→ canonicalQwen/Qwen3.8-Flash-Next) to the model catalog per user request. Verified it is blocked for ONNX conversion and documented findings indocs/blocked-models.md.Findings
Qwen/Qwen3.8-Flash-Next(safetensors, not just the unsloth GGUF repackaging)architectures: ["Qwen4ExpForConditionalGeneration"],model_type: "qwen4_exp"— an experimental preview of the Qwen4 architecture (not Qwen3), combining hybrid linear/sparse attention (Gated DeltaNet + Qwen Sparse Attention), 512-expert MoE, a novel Gated Residual mechanism, and a 51B-param N-gram Embedding table, plus a vision encoder (multimodal).python -m onnxruntime_genai.models.builder(v0.15.1, this repo's pinned version) against the real HF repo inconfig_onlymode — fails immediately becausetransformers5.14.1 (newest available) doesn't recognizeqwen4_expat all (KeyError: 'qwen4_exp'inCONFIG_MAPPING), before onnxruntime-genai's own architecture dispatch table is even reached.microsoft/onnxruntime-genaiconfirms zero occurrences ofqwen4_exp,indexer_budget,hc_lowrank, orngram_vocab_size_baseanywhere in the repo — no partial support exists.Change
Adds a new "Qwen3.8-Flash-Next" section to
docs/blocked-models.md(under Next-Gen Models) following the existing Nemotron 3.5 Lightning / Muse Glimmer template: architecture table, verified dispatch findings, what would unblock it, and recommended alternatives (GGUF via llama.cpp/Unsloth, or Qwen3-14B-Instruct / Qwen2.5-32B-Instruct for native ONNX).No
KnownModels.csentry added — this model stays out of the ONNX-backed catalog until/unless a working conversion path exists.Co-authored-by: Copilot 223556219+Copilot@users.noreply.github.com