mirror of
https://github.com/huggingface/text-generation-inference.git
synced 2025-04-21 23:12:07 +00:00
c1638a56f1
Deepseek V2 is a MoE model from Deepseek. Relevant variations
compared to other models:
- Grouped top-K in expert selection.
- mscale in yarn is calculated using the `mscale` and `mscale_all_dim`
configuration options.
- `mscale_all_dim` is also used in scaling attention softmax.
- Permuting of the query/key representations before applying rotary
embeddings.
- Some projections cannot be sharded (`q_a_proj`, `kv_a_proj_with_mqa`).
So, we need weight loads that supports quantized weights. To this
end `{Weights,WeightLoader}.get_weight` was added.
- The query/key head dimensionality differs from that of the value,
so we need to pad during attention.
- Heads with size 192, needs an extension to our paged attention
fork and we need to ensure that the KV cache is allocated with the
correct size.
- Shared experts.
|
||
|---|---|---|
| .. | ||
| __init__.py | ||
| bloom_modeling.py | ||
| clip.py | ||
| flash_cohere_modeling.py | ||
| flash_dbrx_modeling.py | ||
| flash_deepseek_v2_modeling.py | ||
| flash_gemma2_modeling.py | ||
| flash_gemma_modeling.py | ||
| flash_gpt2_modeling.py | ||
| flash_llama_modeling.py | ||
| flash_mistral_modeling.py | ||
| flash_mixtral_modeling.py | ||
| flash_neox_modeling.py | ||
| flash_pali_gemma_modeling.py | ||
| flash_phi_modeling.py | ||
| flash_qwen2_modeling.py | ||
| flash_rw_modeling.py | ||
| flash_santacoder_modeling.py | ||
| flash_starcoder2_modeling.py | ||
| idefics2.py | ||
| idefics_config.py | ||
| idefics_image_processing.py | ||
| idefics_modeling.py | ||
| idefics_perceiver.py | ||
| idefics_processing.py | ||
| idefics_vision.py | ||
| llava_next.py | ||
| mamba_modeling.py | ||
| mpt_modeling.py | ||
| neox_modeling.py | ||
| opt_modeling.py | ||
| phi_modeling.py | ||
| siglip.py | ||
| t5_modeling.py | ||
| vlm.py | ||