Add mistral3 architecture support for Ministral 3B - #440
Open
insecure-erasure wants to merge 1 commit into
Open
Conversation
insecure-erasure
force-pushed
the
mistral3-support
branch
6 times, most recently
from
April 18, 2026 07:28
c3e0046 to
f0b733f
Compare
- Add mistral3 to TXT_ARCH_LIST - Extend gguf_clip_loader branch to handle mistral3 alongside llama - Fix hardcoded temb shape (131072, 5120) to also accept (131072, 3072) for Ministral 3B hidden_size 3072 - Apply llama_permute and tekken tokenizer reconstruction to mistral3 Fixes city96#439
insecure-erasure
force-pushed
the
mistral3-support
branch
from
April 18, 2026 07:28
f0b733f to
0c43fb7
Compare
|
This is a duplicate of #436 If it’s about code factoring, it would be better to discuss it with the author of that PR no? since for me, the other PR already works fine. |
Author
|
Yes, I agree. I didn't notice the PR. Only checked the issue and started working on the PR right away. I don't have strong opinions about coding styles in a project that doesn't belong to me, so seems fair to let the original author of the PR know about this duplicate. |
ChrisColeTech
added a commit
to ChrisColeTech/ComfyUI-GGUF-Loader
that referenced
this pull request
Aug 12, 2026
Cherry-pick production fixes from city96/ComfyUI-GGUF open PRs: - city96#472 dequant device-constant cache (major LTX/sampling speedup) - city96#433 IQ1/IQ2/IQ3 torch dequant (extra TE quants) - city96#470 QK-norm .scale→.weight (silent NaN/black Flux-compat) - city96#467 dequant bare nn.Parameters (LTX learnable_registers) - city96#392 lumina2/zimage pad token shape fix - city96#456/city96#468 GGMLTensor dtype + dequantize() for core cast path - city96#461 WeightAdapter-aware move_patch_to_device - city96#469 force_patch on partial load/unload - city96#440/city96#436 mistral3 TE, city96#438 qwen35, qwen2 allowlist - city96#473 partial: Qwen3-VL deepstack mmproj map for MiniMax H3 TE Skipped mega/draft rewrites (city96#459, full city96#473 LazyGGUFReader, city96#336 Triton). See PR_BACKPORT.md for the full matrix.
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
2 participants
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.
Fixes #439
Adds support for
mistral3architecture (official K-quant GGUFs frommistralai/Ministral-3-3B-Instruct-2512-GGUF). Without this, loading them raisesValueError: Unexpected text model architecture type: 'mistral3'.Also fixes the hardcoded shape
(131072, 5120)ingguf_tekken_tokenizer_loaderto also accept(131072, 3072)forMinistral 3B. This one affects any GGUF of this model regardless of origin.
Tested with ERNIE-Image Turbo (NVFP4) + official Q5_K_M on 8 GB VRAM.