ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

model : Granite MoE shared (#13269)

Browse source 
* feat: Add GGUF conversion for granitemoeshared

Branch: GraniteMoEShared

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* feat: hparam and arch plumbing for granitemoeshared

Branch: GraniteMoEShared

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* fix: Split MoE fused tensors for shared experts in conversion

Branch: GraniteMoEShared

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* feat: First WIP cut at model arch in cpp

The hparam and architecture plumbing should be correct, but the
implementation of the shared experts seems to still be broken.

Branch: GraniteMoEShared

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* fix: Cleaner (maybe more correct?) splitting for gate/up

Branch: GraniteMoEShared

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* fix: Fix the input to the shared experts

I had misread that the shared experts take the inputs _before_ the standard
MoE layer and was feeding the output of the MoE to the shared experts.

Branch: GraniteMoEShared

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* fix: Avoid architecture-specific checks for Granite MoE Shared

This is a cleaner way that will allow more flexibility in architecture
strings going forward.

Branch: GraniteMoEShared

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* refactor: Split granite architectures out of llm_build_llama

This helps de-clutter the llama-family graph construction and allows
granite to diverge further (in preparation for Granite 4).

NOTE: I removed the granite scale factors from llm_build_deci because they
appear to only be there as copy-paste from llm_build_llama. The HF config
does not seem to set those values:
https://huggingface.co/Deci/DeciLM-7B/blob/main/config.json

Branch: GraniteMoEShared

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* fix: Fix compiler warning about uninitialized inp_pos

This should not have been reachable, but it warns on some compliers

Branch: GraniteMoEShared

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* fix: Consoladate GraniteMoEShared into GraniteMoE for conversion

Branch: GraniteMoEShared

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

* fix: Consolidate GraniteMoEShared into GraniteMoE on the c++ side

Branch: GraniteMoEShared

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>

---------

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
This commit is contained in:
Gabe Goodhart 2025-05-13 07:12:01 -06:00 • committed by GitHub
parent 1e2809bc4b
commit d590cd4c24
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
5 changed files with 235 additions and 35 deletions
Download patch file Download diff file Expand all files Collapse all files

22
convert_hf_to_gguf.py
View file

@ -5746,11 +5746,20 @@ class GraniteModel(LlamaModel):
logger.info("gguf: (granite) logits_scale = %s", logits_scale)
@ModelBase.register("GraniteMoeForCausalLM")
@ModelBase.register("GraniteMoeForCausalLM", "GraniteMoeSharedForCausalLM")
class GraniteMoeModel(GraniteModel):
"""Conversion for IBM's GraniteMoeForCausalLM"""
model_arch = gguf.MODEL_ARCH.GRANITE_MOE
def set_gguf_parameters(self):
"""GraniteMoeShared uses GraniteMoe parameters plus the following:
- shared_intermediate_size
"""
super().set_gguf_parameters()
if shared_feed_forward_length := self.hparams.get("shared_intermediate_size"):
self.gguf_writer.add_expert_shared_feed_forward_length(shared_feed_forward_length)
logger.info("gguf: (granitemoeshared) shared_feed_forward_length = %s", shared_feed_forward_length)
def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
"""In modeling_granitemoe, the JetMoe implementation of parallel experts
is used. This essentially merges w1 and w3 into a single tensor with 2x
@ -5761,12 +5770,21 @@ class GraniteMoeModel(GraniteModel):
if name.endswith("block_sparse_moe.input_linear.weight"):
ffn_dim = self.hparams["intermediate_size"]
assert data_torch.shape[-2] == 2 * ffn_dim, "Merged FFN tensor size must be 2 * intermediate_size"
gate, up = data_torch[..., :ffn_dim, :], data_torch[..., ffn_dim:, :]
gate, up = data_torch.split(ffn_dim, dim=-2)
return [
(self.format_tensor_name(gguf.MODEL_TENSOR.FFN_GATE_EXP, bid), gate),
(self.format_tensor_name(gguf.MODEL_TENSOR.FFN_UP_EXP, bid), up),
]
if name.endswith("shared_mlp.input_linear.weight"):
ffn_dim = self.hparams["shared_intermediate_size"]
assert data_torch.shape[-2] == 2 * ffn_dim, "Merged FFN tensor size must be 2 * shared_intermediate_size"
gate, up = data_torch.split(ffn_dim, dim=-2)
return [
(self.format_tensor_name(gguf.MODEL_TENSOR.FFN_GATE_SHEXP, bid), gate),
(self.format_tensor_name(gguf.MODEL_TENSOR.FFN_UP_SHEXP, bid), up),
]
return super().modify_tensors(data_torch, name, bid)