Issues

• Gpt-Oss-20B follow-up - Yarn embeddings in FMS way

  foundation-model-stack/foundation-model-stack#511 · opened on Mar 3, 2026

• Use search instead of match in _hf_to_fms_rope() adapter

  foundation-model-stack/foundation-model-stack#491 · opened on Nov 21, 2025

• Gpt-Oss-20B follow-up - Implement TPMultiheadAttention support

  foundation-model-stack/foundation-model-stack#480 · opened on Oct 17, 2025

• Gpt-Oss-20B follow-up - Add HF equivalency tests

  foundation-model-stack/foundation-model-stack#479 · opened on Oct 15, 2025

• Gpt-Oss-20B follow-up - Model unfuse weights

  foundation-model-stack/foundation-model-stack#478 · opened on Oct 15, 2025

• Gpt-Oss-20B follow-up - Add optmizations to attention and MoE code.

  foundation-model-stack/foundation-model-stack#477 · opened on Oct 15, 2025

• Add support to openai/gpt-oss-20b

  foundation-model-stack/foundation-model-stack#461 · opened on Aug 20, 2025

• Migrate to pyproject.toml based packaging of FMS

  foundation-model-stack/foundation-model-stack#450 · opened on Jul 14, 2025

• [serialization] no guarantee that dictionary has to or device

  Bug

  foundation-model-stack/foundation-model-stack#443 · opened on Jul 7, 2025
  1

• Add unit tests for the tekken tokenizer class

  foundation-model-stack/foundation-model-stack#439 · opened on Jul 2, 2025

• Port mistralai/Mistral-Small-3.1-24B-Instruct-2503 to FMS

  foundation-model-stack/foundation-model-stack#428 · opened on Jun 23, 2025
  1

• Infer GPTQ model from hf model config

  foundation-model-stack/foundation-model-stack#392 · opened on Apr 8, 2025

• Issue loading models from HF Hub on CPU

  foundation-model-stack/foundation-model-stack#376 · opened on Jan 16, 2025
  2

• Variable naming

  foundation-model-stack/foundation-model-stack#350 · opened on Oct 9, 2024

• new llama variants cannot be handled through adapters as the head size is not supported

  foundation-model-stack/foundation-model-stack#344 · opened on Sep 26, 2024

• GPTQ fails with cuda-old kernels

  foundation-model-stack/foundation-model-stack#337 · opened on Sep 12, 2024

• ghstack conflicts with new DCO checks

  foundation-model-stack/foundation-model-stack#323 · opened on Aug 13, 2024

• Roberta loading from HF

  foundation-model-stack/foundation-model-stack#300 · opened on Jul 10, 2024

• Mixtral Loading from safetensors

  foundation-model-stack/foundation-model-stack#299 · opened on Jul 10, 2024

• QKV double-initialization

  foundation-model-stack/foundation-model-stack#298 · opened on Jul 9, 2024

• torch.export integration with FMS

  foundation-model-stack/foundation-model-stack#293 · opened on Jun 27, 2024
  3

• Would Be Helpful to Set A & B When Calling trunc_normal_ in Initialization

  foundation-model-stack/foundation-model-stack#289 · opened on Jun 21, 2024

• upgrade lm eval script to allow latest version of harness

  foundation-model-stack/foundation-model-stack#276 · opened on May 30, 2024

• AssertionError: Hidden dim must be divisible by world size

  foundation-model-stack/foundation-model-stack#274 · opened on May 27, 2024
  3

• [WIP] Rotated Quantization for Accuracy in Matrix Multiplication

  foundation-model-stack/foundation-model-stack#271 · opened on May 23, 2024