no_rope_layer_interval is declared but never applied, and SmolLM3's own config sets it
维护者通常 1 天内回复
评估
这个 Issue 还没有评估数据。
描述
ModelArgs declares no_rope_layer_interval, and examples/models/smollm3/3b_config.json sets it to 4, but nothing on the examples/models/llama export path reads it. On main (e4576d0) the name appears exactly once across the 32 Python files of that path:
examples/models/llama/model_args.py:119: no_rope_layer_interval: Optional[int] = (
__post_init__ does not touch it, and Attention.forward applies rope unconditionally (attention.py:529, attention.py:563) with no layer-index test. The two places that do honour it are backends/mlx/llm/et_attention.py:111 and examples/qualcomm/oss_scripts/llama/model/static_llama.py:288.
So a model that alternates RoPE and NoPE layers exports without complaint and comes out with rope on every layer.
Reproducing
examples/models/smollm3/3b_config.json and examples/models/smollm3/convert_weights.py are both in the tree, so SmolLM3-3B looks ready to export. Converting its weights and exporting through export_llm with that params json (XNNPACK, 8da4w + 8-bit embedding, 1852.2 MB) gives a .pte that loads and runs, then does this:
prompt: <|im_start|>user\nWhat is the capital of France?<|im_end|>\n<|im_start|>assistant\n
output: Okay, so you want to know what what what what what what what what what what what
what what what what what what what what what what what what what what what what
prompt: <|im_start|>user\nWhat is 17 times 4?<|im_end|>\n<|im_start|>assistant\n
output: I think you are looking for an answer that can be given that can be be be be be
determ determ determ determ determ determ determ determ determ determ#ae determ#
It starts as English and collapses, which is what a positional-encoding mismatch looks like: the first few tokens are fine and the drift grows with position.
What would help
Either implement the interval on this path, or reject a params json that sets a field the chosen backend does not implement. The silent version is the expensive one — the export succeeds, the file is the right size, and the model is simply a different model.
Measured with the executorch 1.4.0 wheel; the code paths quoted above are byte-identical on main at e4576d0.
cc @larryliu0820 @mergennachin @cccclai @helunwencser @jackzhxng @digantdesai
- 主要语言
- Python
- 星标
- 5.1k
- 派生
- 1.2k
- 平均合并
- 1 天 23 小时
- 30 天内合并 PR
- 713
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
pytorch/executorch 的其他 Issue
-
Portable remainder kernel: wrong sign for integers and float64 rounded to float可能已有人在做 @JakeStevens 于 4 天前认领。 未关闭bug module: kernels triaged
难度 2/5 1-3 小时 新手友好度 82/100
pytorch/executorch#23682 · 已指派 2 人 ·
维护者通常 1 天内回复
-
enhancement triaged
难度 2/5 半天 新手友好度 68/100
pytorch/executorch#21640 ·
维护者通常 1 天内回复
-
bug module: microcontrollers partner: arm
难度 4/5 3-5 天 新手友好度 38/100
pytorch/executorch#23812 · 2 条评论 · 1 个 reaction ·
维护者通常 1 天内回复
-
[muse-glimmer] interpolate_pos_emb returns corrupted bf16 values on CPU and can segfault可能已有人在做 @JakeStevens 于 2 天前认领。 未关闭bug module: examples triaged
pytorch/executorch#23797 · 2 条评论 · 已指派 1 人 ·
维护者通常 1 天内回复
-
macOS Python wheel _portable_lib has non-relocatable libtorch rpath可能已有人在做 @JacobSzwejbka 于 1 天前认领。 未关闭
pytorch/executorch#23742 · 1 条评论 · 已指派 2 人 ·
维护者通常 1 天内回复
查看 pytorch/executorch 的全部 Issue
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 65/100
-
难度 2/5 1-3 小时 新手友好度 72/100
UKGovernmentBEIS/control-arena#917 ·
维护者通常 1 天内回复
-
难度 1/5 1 小时以内 新手友好度 88/100
-
难度 2/5 1-3 小时 新手友好度 82/100
debpalash/VoiceStudio#2937 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 72/100
aws-samples/sample-per-model-bedrock#261 ·
维护者通常 1 天内回复