Add inkling2 #50682

guarin merged 69 commits into main from add-inkling2
guarin
molbap draft
40d43055
molbap simplify processor a bit
f3dd8930
molbap reduce modular, 4d pixel values and fix mappings
6f3b9fb8
molbap small iteration
35541104
guarin ternary
b93a00ce
guarin refactor decoder layer forward
d66306fd
guarin refactor block_attention_residual
1c3719bc
guarin add Inkling2BlockAttentionResidual
16676a3b
guarin keep vision encoder in block order
61557170
molbap Merge pull request #3 from huggingface/update-modular
b793ec0a
molbap remove copy?
6ae08c0b
molbap docs + checkerboard testing
82caa46a
molbap fix router bias precision, parallel plans and log-scaling positions
578c68cb
molbap remove shared experts and unused config options
15786c62
molbap remove host syncs from the vision path
55f23276
molbap address review comments on config and modules
8348c659
molbap share test overrides and move the vision parity check out
7066cea0
molbap add a processor test
9257f858
molbap better naming
602c1e1c
guarin Merge branch 'main' into add-inkling2
09c75518
guarin torch.Tensor
dddb06d5
guarin Merge branch 'main' into fast-load
43e78e6c
guarin [inkling2] load the NVFP4 / MXFP8 experts through the quantization plan
fccda543
guarin add Inkling2ScaledL2Clip
cb959ef5
guarin add linear
74cf4ba2
guarin add comment
8aec11b0
guarin update router/moe
29be87a3
guarin fix is_concatenated
88e42122
guarin add Inkling2Softcap
d8ee70b2
guarin remove n_shared_experts
f16c9602
guarin inherit vision config
dd0fa4fd
guarin cleanup image processor
636cecad
guarin processor image token handling
2db59081
guarin split attention classes
b4b6d69f
guarin use norm layer
6be2a57f
guarin update docs
74573ed2
guarin update docstrings
1e767543
guarin enable sdpa for vision tower
3df17228
guarin small improvements
f8bcef31
guarin update config
a97a1515
guarin hardcode missing config values
f461941d
guarin keep block states as tensors
e800588e
guarin update tests
2ed6bd03
guarin [inkling2] keep the convolution weight renaming off `weight_proj`
76c5fd94
guarin Merge branch 'main' into add-inkling2
617ac971
guarin Merge branch 'add-inkling2' into fast-load
968c8791
guarin Merge pull request #4 from huggingface/fast-load
2ad0e4b4
guarin [inkling2] quantization config as glob patterns
6a1e78e3
guarin fix export
05ca9c44
guarin fix bos token
b0580b7f
guarin add tokenizer
a91fc50d
guarin [inkling2] move block_states to each layer's device (device_map)
cc86252e
guarin Merge remote-tracking branch 'upstream/main' into add-inkling2
a38bf7c6
guarin [inkling2] default pad_token to <|endoftext|> (checkpoints don't set …
4d0429be
guarin [inkling2] small integration test: pin logits measured on CUDA with t…
325bf24b
guarin Revert "[inkling2] default pad_token to <|endoftext|> (checkpoints do…
6b00849a
guarin remove tokenizer
cb8e5fa4
guarin small improvement
d687f4be
guarin Merge remote-tracking branch 'upstream/main' into add-inkling2
719dbaa6
guarin update docs
c7f60cf9
guarin Fix onnx export
c1bbc7c3
guarin add CI expectations
832708f7
guarin Merge branch 'main' into add-inkling2
0237409d
guarin small fixes
8e79c4e3
guarin update docs
92b568c9
guarin make it work
226521c5
guarin update docs
1b2f6259
vasqu vasqu added New model
vasqu
vasqu approved these changes on 2026-11-12
github-actions
vasqu
vasqu commented on 2026-11-12
vasqu
vasqu commented on 2026-11-12
vasqu
vasqu commented on 2026-11-12
guarin cleanup
64ccab4b
guarin rename
a8741344
guarin
HuggingFaceDocBuilderDev
github-actions
guarin
guarin guarin merged 109a7d19 into main 58 minutes ago
guarin guarin deleted the add-inkling2 branch 58 minutes ago

Login to write a write a comment.

Login via GitHub

Reviewers
Assignees
No one assigned
Labels
Milestone