model : Granite MoE shared (#13269)

gabe-l-hart · web-flow · commit d590cd4c244e · 2025-05-13T15:12:01.000+02:00
* 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>
diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py
@@ -5746,11 +5746,20 @@ def set_gguf_parameters(self):
             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 @@ def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iter
         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)
 
 
diff --git a/gguf-py/gguf/constants.py b/gguf-py/gguf/constants.py
@@ -1905,6 +1905,9 @@ class MODEL_TENSOR(IntEnum):
         MODEL_TENSOR.FFN_GATE_EXP,
         MODEL_TENSOR.FFN_DOWN_EXP,
         MODEL_TENSOR.FFN_UP_EXP,
+        MODEL_TENSOR.FFN_GATE_SHEXP,
+        MODEL_TENSOR.FFN_UP_SHEXP,
+        MODEL_TENSOR.FFN_DOWN_SHEXP,
     ],
     MODEL_ARCH.CHAMELEON: [
         MODEL_TENSOR.TOKEN_EMBD,
diff --git a/gguf-py/gguf/tensor_mapping.py b/gguf-py/gguf/tensor_mapping.py
@@ -428,6 +428,7 @@ class TensorNameMap:
             "model.layers.{bid}.mlp.shared_expert.down_proj",  # qwen2moe
             "model.layers.{bid}.mlp.shared_experts.down_proj", # deepseek deepseek2
             "language_model.model.layers.{bid}.feed_forward.shared_expert.down_proj", # llama4
+            "model.layers.{bid}.shared_mlp.output_linear",     # granitemoe
         ),
 
         MODEL_TENSOR.ATTN_Q_NORM: (
diff --git a/src/llama-arch.cpp b/src/llama-arch.cpp
@@ -1481,6 +1481,9 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
             { LLM_TENSOR_FFN_GATE_EXPS,   "blk.%d.ffn_gate_exps" },
             { LLM_TENSOR_FFN_DOWN_EXPS,   "blk.%d.ffn_down_exps" },
             { LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },
+            { LLM_TENSOR_FFN_GATE_SHEXP,  "blk.%d.ffn_gate_shexp" },
+            { LLM_TENSOR_FFN_DOWN_SHEXP,  "blk.%d.ffn_down_shexp" },
+            { LLM_TENSOR_FFN_UP_SHEXP,    "blk.%d.ffn_up_shexp" },
         },
     },
     {
diff --git a/src/llama-model.cpp b/src/llama-model.cpp
