DevVexus 9a7570587c convert : write explicit recurrent_layers for Qwen3-Next / Qwen3.5 (#28208) ... Problem - Loader prefers `<arch>.attention.recurrent_layers`, falls back to `full_attention_interval` if missing - Converter only ever writes the interval. gguf-py has no constant/writer for the array - Interval can only describe evenly spaced full-attention layers. Any non-uniform `layer_types` gets reconstructed wrong - No error, no warning. Model loads, runs, wrong layers get wrong ops. Full-attn layers marked recurrent lose their KV cache - Every published Qwen3.5 checkpoint is uniform so nobody's hit it yet Repro 12 layers, periods 4/3/5: layer: 0 1 2 3 4 5 6 7 8 9 10 11 actual: L L L F L L F L L L L F loader: L L L F L L L F L L L F ^ ^ Layer 6 is full attn, loaded as recurrent. Layer 7 the reverse. 52-layer non-uniform stack: 15/52 mis-typed. Fix - `constants.py`: add `Keys.Attention.RECURRENT_LAYERS` (name already registered in llama-arch.cpp) - `gguf_writer.py`: add `add_recurrent_layers()`, same shape as `add_rope_pattern()` - `conversion/qwen.py`: emit array from `layer_types` in `Qwen3NextModel.set_gguf_parameters` (covers 3-Next, 3.5, 3.5-MoE) Notes - Array is padded with `false` for MTP blocks. `get_key_or_arr` checks length against `n_layer_all`, which includes MTP. Matches the fallback's `i < n_layer()` guard - Interval is still written. Old builds only understand the interval - `layer_types` length != `num_hidden_layers` now raises in converter instead of producing a GGUF that fails at load Tested - End-to-end on a 62-layer non-uniform Qwen3.8-27B (2 linear layers removed). Loader reads the array, 62 blocks, 0 mismatches. Without fix: interval fallback, mis-typed - MTP padding NOT tested on a real MTP model. Reasoned from qwen35.cpp + get_key_or_arr. Would appreciate a check Co-authored-by: Claude Opus 5 <noreply@anthropic.com>		2026-09-07 10:12:50 +03:00
..
__init__.py	[Model] Support for Spark2_5ForCausalLM implementation (#27868)	2026-09-06 17:43:58 +02:00
afmoe.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
arctic.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
baichuan.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
bailingmoe.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
bailingmoe3.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
base.py	convert : add --fuse-qkv flag to fuse Q/K/V into QKV during HF-to-GGUF conversion (#22780)	2026-09-07 00:47:05 +08:00
bert.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
bitnet.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
bloom.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
chameleon.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
chatglm.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
codeshell.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
cogvlm.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
command_r.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
dbrx.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
deci.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
deepseek.py	convert : skip bias_vl tensor in DeepSeek-V4 DSpark conversion (#28294)	2026-09-03 10:37:23 +03:00
dots1.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
dots3.py	mtmd: support dots3-note vision+audio (#27524)	2026-08-22 10:35:50 +02:00
dotsocr.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
dream.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
ernie.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
exaone.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
falcon.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
falcon_h1.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
gemma.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
glm.py	convert: fix GLM regression in index_tensors (#27655)	2026-08-24 13:21:00 +03:00
gpt2.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
gpt_oss.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
gptneox.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
granite.py	model : GraniteSWAForCausalLM / GraniteMoeSWAForCausalLM (#25505)	2026-08-19 16:53:31 +02:00
grok.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
grovemoe.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
hunyuan.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
hy_v4.py	model : add Tencent Hy 4 (hy_v4) preview architecture support (#28127)	2026-09-04 14:31:36 +02:00
internlm.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
internvl.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
jais.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
jamba.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
januspro.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
kimi_k3.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
kimi_linear.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
kimivl.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
laguna.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
lfm2.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
lighton_ocr.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
llada.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
llama.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
llama4.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
llava.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
maincoder.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
mamba.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
mellum.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
mimo.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
minicpm.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
minimax.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
mistral.py	convert : fix conversion for Mistral-Medium-3.5-128B (#24268)	2026-06-07 21:41:39 +02:00
mistral3.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
mpt.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
muse_glimmer.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
nanbeige.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
nemotron.py	model: add NVIDIA Nemotron-3-Puzzle-75B-A9B (NemotronHPuzzle) support (#25444)	2026-09-03 08:53:08 +02:00
olmo.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
openelm.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
orion.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
pangu.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
phi.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
pixtral.py	Refactor: convert_hf_to_gguf.py (#17114)	2026-05-15 15:18:12 +02:00
plamo.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
plm.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
pockettts.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
qwen.py	convert : write explicit recurrent_layers for Qwen3-Next / Qwen3.5 (#28208)	2026-09-07 10:12:50 +03:00
qwen3tts.py	mtmd: Fix Qwen3-tts-0.6b (#28231)	2026-09-02 12:46:16 +02:00
qwen3vl.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
qwen4exp.py	model: add Qwen3.8-Flash-Next (qwen4exp) (#27742)	2026-08-27 21:32:31 +02:00
qwenvl.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
refact.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
rwkv.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
sarashina2.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
smallthinker.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
smolvlm.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
spark2_5.py	[Model] Support for Spark2_5ForCausalLM implementation (#27868)	2026-09-06 17:43:58 +02:00
stablelm.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
starcoder.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
step3.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
t5.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
talkie.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
ultravox.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
wavtokenizer.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
xverse.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00
youtuvl.py	convert: add @ModelBase.example (#27208)	2026-08-17 10:15:11 +02:00