[Docs] [V1] [Hybrid] Add new documentation re: contributing mamba-based models

docs/contributing/model/basic.md

@@ -121,3 +121,29 @@ To support a model with interleaving sliding windows, we need to take care of th
 - In the modeling code, parse the correct sliding window value for every layer, and pass it to the attention layer's `per_layer_sliding_window` argument. For reference, check [this line](https://github.com/vllm-project/vllm/blob/996357e4808ca5eab97d4c97c7d25b3073f46aab/vllm/model_executor/models/llama.py#L171).
 
 With these two steps, interleave sliding windows should work with the model.
+
+### How to support models that use Mamba?
+
+We consider 3 different scenarios:
+
+1. Models that use Mamba layers (either Mamba-1 or Mamba-2) but do not use attention layers.
+2. Models that combine Mamba layers (either Mamba-1 or Mamba-2) together with attention layers.
+3. Models that combine Mamba-like mechanisms (e.g., Linear Attention, ShortConv) together with attention layers.
+
+For case (1), we recommend looking at the implementation of [`MambaForCausalLM`](gh-file:vllm/model_executor/models/mamba.py) (for Mamba-1) or [`Mamba2ForCausalLM`](gh-file:vllm/model_executor/models/mamba2.py) (for Mamba-2) as a reference.
+The model should inherit protocol `IsAttentionFree` and also implement class methods `get_mamba_state_dtype_from_config` and `get_mamba_state_shape_from_config` to calculate the state shapes and data dtypes from the config.
+For the mamba layers themselves, please use the [`MambaMixer`](gh-file:vllm/model_executor/layers/mamba/mamba_mixer.py) (for Mamba-1) or [`MambaMixer2`](gh-file:vllm/model_executor/layers/mamba/mamba_mixer2.py) (for Mamba-2) classes.
+The model should also be added to the [`MODELS_CONFIG_MAP`](gh-file:vllm/model_executor/models/config.py#L419-L421) dictionary to ensure that the runtime defaults are optimized.
+
+For case (2), we recommend using as a reference the implementation of [`JambaForCausalLM`](gh-file:vllm/model_executor/models/jamba.py) (for an example of a model that uses Mamba-1 and attention together) or [`BambaForCausalLM`](gh-file:vllm/model_executor/models/bamba.py) (for an example of a model that uses Mamba-2 and attention together).
+These models should follow the same instructions as case (1), but they should inherit protocol `IsHybrid` (instead of `IsAttentionFree`) and it is *not* necessary to add them to the `MODELS_CONFIG_MAP` (their runtime defaults will be inferred from the protocol).
+
+For case (3), we recommend looking at the implementation of [`MiniMaxText01ForCausalLM`](gh-file:vllm/model_executor/models/minimax_text_01.py) or [`Lfm2ForCausalLM`](gh-file:vllm/model_executor/models/lfm2.py) as a reference, which use custom "mamba-like" layers [`MiniMaxText01LinearAttention`](gh-file:vllm/model_executor/models/minimax_text_01.py#L287) and [`ShortConv`](gh-file:vllm/model_executor/layers/mamba/short_conv.py) respectively.
+Please follow the same guidelines as case (2) for implementing these models.
+We use "mamba-like" to refer to layers that posses a state that is updated in-place, rather than being conconcated-to (like KV cache for attention).
+For implementing new custom mamba-like layers, one should inherit from `MambaBase` and implement the methods `get_state_dtype`, `get_state_shape` to calculate the data types and state shapes at runtime, as well as `mamba_type` and `get_attn_backend`.
+It is also necessary to implement the "attention meta-data" class which handles the meta-data that is common across all layers.
+Please see [`LinearAttentionMetadata`](gh-file:vllm/v1/attention/backends/linear_attn.py) or [`ShortConvAttentionMetadata`](gh-file:v1/attention/backends/short_conv_attn.py) for examples of this.
+Finally, if one wants to support torch compile and CUDA graphs, it necessary to wrap the call to the mamba-like layer inside a custom op.
+Please see examples for [`linear-attention`](gh-file:vllm/model_executor/models/minimax_text_01.py#L1402-L1431) and [`short-conv`](gh-file:vllm/model_executor/layers/mamba/short_conv.py#L244-L270) respectively.
+The new custom op should then be added to the list [here](gh-file:vllm/config/compilation.py#L335-L343) to ensure that piecewise CUDA graphs works as intended.

docs/usage/v1_guide.md

@@ -107,16 +107,14 @@ to enable simultaneous generation and embedding using the same engine instance i
 #### Mamba Models
 
 Models using selective state-space mechanisms instead of standard transformer attention are supported.
-Models that use Mamba-2 and Mamba-1 layers (e.g., `Mamba2ForCausalLM`, `MambaForCausalLM`) are supported.
-Please note that prefix caching is not yet supported for these models.
+Models that use Mamba-2 and Mamba-1 layers (e.g., `Mamba2ForCausalLM`, `MambaForCausalLM`,`FalconMambaForCausalLM`) are supported.
 
-Models that combine Mamba-2 and Mamba-1 layers with standard attention layers are also supported (e.g., `BambaForCausalLM`,
+Hybrid models that combine Mamba-2 and Mamba-1 layers with standard attention layers are also supported (e.g., `BambaForCausalLM`,
 `Zamba2ForCausalLM`, `NemotronHForCausalLM`, `FalconH1ForCausalLM` and `GraniteMoeHybridForCausalLM`, `JambaForCausalLM`).
-Please note that prefix caching is not yet supported for these models.
 
-Hybrid models with mechanisms different to Mamba are also supported (e.g, `MiniMaxText01ForCausalLM`, `MiniMaxM1ForCausalLM`).
-Please note that prefix caching is not yet supported for these models.
-It is also necessary to enforce eager mode for these models in V1.
+Hybrid models with mechanisms different to Mamba are also supported (e.g, `MiniMaxText01ForCausalLM`, `MiniMaxM1ForCausalLM`, `Lfm2ForCausalLM`).
+
+Please note that prefix caching is not yet supported for any of the above models.
 
 #### Encoder-Decoder Models