arcee-ai/mergekit v0.1.5
arcee-ai/mergekit
Captured source
source ↗published Oct 20, 2026seen 3dcaptured 3dhttp 200method plain
v0.1.5
Repository: arcee-ai/mergekit
Tag: v0.1.5
Published: 2026-10-20T11:48:50Z
Prerelease: no
Release notes:
What's Changed
- Fix auto-architecture for MoE models by @cg123 in https://github.com/arcee-ai/mergekit/pull/647
- Add AFMoE support by @cg123 in https://github.com/arcee-ai/mergekit/pull/648
- Add ApertusForCausalLM by @cg123 in https://github.com/arcee-ai/mergekit/pull/655
- Allow tokensurgeon to attempt auto architecture by @cg123 in https://github.com/arcee-ai/mergekit/pull/656
- Bug fix: Filter tokens not in the embedding (tokensurgeon) by @xz-liu in https://github.com/arcee-ai/mergekit/pull/657
- Add RAM and tensor-local RAM+ variant by @cg123 in https://github.com/arcee-ai/mergekit/pull/662
- Update tokenizer merging documentation by @cg123 in https://github.com/arcee-ai/mergekit/pull/663
- Add GLM4 MOE models support by @pengdurice in https://github.com/arcee-ai/mergekit/pull/658
- Add Qwen3-VL support by @pengdurice in https://github.com/arcee-ai/mergekit/pull/665
- Add Qwen3-VL dense models by @pengdurice in https://github.com/arcee-ai/mergekit/pull/670
- Add IBM Granite architecture support by @Bhavyashah20 in https://github.com/arcee-ai/mergekit/pull/683
- Fix TypeError when loading model in MoE router without bitsandbytes by @umran666 in https://github.com/arcee-ai/mergekit/pull/690
- Update to Transformers v5 by @cg123 in https://github.com/arcee-ai/mergekit/pull/692
- Bump copyright date by @cg123 in https://github.com/arcee-ai/mergekit/pull/694
- Add arXiv link for Karcher Mean paper by @neoheartbeats in https://github.com/arcee-ai/mergekit/pull/673
- Add missing
norm.weightparameters tocohere.jsonby @jukofyork in https://github.com/arcee-ai/mergekit/pull/674 - Relax dependencies + switch to ruff by @cg123 in https://github.com/arcee-ai/mergekit/pull/705
- Preserve existing tokenizer metadata by @NeonBohdan in https://github.com/arcee-ai/mergekit/pull/703
- fix(moe/deepseek): create the output directory before writing config.json (#697) by @Anai-Guo in https://github.com/arcee-ai/mergekit/pull/701
- Adding missing q_norm/k_norm weights to olmo2.json by @armsp in https://github.com/arcee-ai/mergekit/pull/699
- Fix ENAMETOOLONG crash when chat_template is an inline Jinja string by @javierdejesusda in https://github.com/arcee-ai/mergekit/pull/698
- New merge method API by @cg123 in https://github.com/arcee-ai/mergekit/pull/706
- Remove vllm extra by @cg123 in https://github.com/arcee-ai/mergekit/pull/717
- Fix magnitude pruning for small tensors in TIES merges by @tomaarsen in https://github.com/arcee-ai/mergekit/pull/716
New Contributors
- @xz-liu made their first contribution in https://github.com/arcee-ai/mergekit/pull/657
- @pengdurice made their first contribution in https://github.com/arcee-ai/mergekit/pull/658
- @Bhavyashah20 made their first contribution in https://github.com/arcee-ai/mergekit/pull/683
- @umran666 made their first contribution in https://github.com/arcee-ai/mergekit/pull/690
- @neoheartbeats made their first contribution in https://github.com/arcee-ai/mergekit/pull/673
- @NeonBohdan made their first contribution in https://github.com/arcee-ai/mergekit/pull/703
- @Anai-Guo made their first contribution in https://github.com/arcee-ai/mergekit/pull/701
- @armsp made their first contribution in https://github.com/arcee-ai/mergekit/pull/699
- @javierdejesusda made their first contribution in https://github.com/arcee-ai/mergekit/pull/698
- @tomaarsen made their first contribution in https://github.com/arcee-ai/mergekit/pull/716
Full Changelog: https://github.com/arcee-ai/mergekit/compare/v0.1.4...v0.1.5