model: support phimoe

#9119

Author: phymbert

convert_hf_to_gguf.py

@@ -2564,6 +2564,62 @@ def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
         yield (self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FACTORS_LONG), torch.tensor(long_factors, dtype=torch.float32))
         yield (self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FACTORS_SHORT), torch.tensor(short_factors, dtype=torch.float32))
 
+@Model.register("PhiMoEForCausalLM")
+class PhiMoeModel(Phi3MiniModel):
+    model_arch = gguf.MODEL_ARCH.PHIMOE
+
+    _experts: list[dict[str, Tensor]] | None = None
+
+    def set_gguf_parameters(self):
+        super().set_gguf_parameters()
+        self.gguf_writer.add_expert_used_count(self.hparams["num_experts_per_tok"])
+        self.gguf_writer.add_expert_count(self.hparams["num_local_experts"])
+
+    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
+        # process the experts separately
+        if name.find("block_sparse_moe.experts") != -1:
+            n_experts = self.hparams["num_local_experts"]
+            assert bid is not None
+
+            if self._experts is None:
+                self._experts = [{} for _ in range(self.block_count)]
+
+            self._experts[bid][name] = data_torch
+
+            if len(self._experts[bid]) >= n_experts * 3:
+                tensors: list[tuple[str, Tensor]] = []
+
+                # merge the experts into a single 3d tensor
+                for w_name in ["w1", "w2", "w3"]:
+                    datas: list[Tensor] = []
+
+                    for xid in range(n_experts):
+                        ename = f"model.layers.{bid}.block_sparse_moe.experts.{xid}.{w_name}.weight"
+                        datas.append(self._experts[bid][ename])
+                        del self._experts[bid][ename]
+
+                    data_torch = torch.stack(datas, dim=0)
+
+                    merged_name = f"model.layers.{bid}.block_sparse_moe.experts.{w_name}.weight"
+
+                    new_name = self.map_tensor_name(merged_name)
+
+                    tensors.append((new_name, data_torch))
+                return tensors
+            else:
+                return []
+
+        return [(self.map_tensor_name(name), data_torch)]
+
+    def prepare_tensors(self):
+        super().prepare_tensors()
+
+        if self._experts is not None:
+            # flatten `list[dict[str, Tensor]]` into `list[str]`
+            experts = [k for d in self._experts for k in d.keys()]
+            if len(experts) > 0:
+                raise ValueError(f"Unprocessed experts: {experts}")
+
 
 @Model.register("PlamoForCausalLM")
 class PlamoModel(Model):

docs/development/HOWTO-add-model.md

@@ -28,7 +28,7 @@ The required steps to implement for an HF model are:
 ```python
 @Model.register("MyModelForCausalLM")
 class MyModel(Model):
-    model_arch = gguf.MODEL_ARCH.GROK
+    model_arch = gguf.MODEL_ARCH.MYMODEL
 ```
 
 2. Define the layout of the GGUF tensors in [constants.py](/gguf-py/gguf/constants.py)
@@ -79,14 +79,14 @@ Depending on the model configuration, tokenizer, code and tensors layout, you wi
 - `Model#set_vocab`
 - `Model#write_tensors`
 
-NOTE: Tensor names must end with `.weight` suffix, that is the convention and several tools like `quantize` expect this to proceed the weights.
+NOTE: Tensor names must end with `.weight` or `.bias` suffixes, that is the convention and several tools like `quantize` expect this to proceed the weights.
 
 ### 2. Define the model architecture in `llama.cpp`
 
 The model params and tensors layout must be defined in `llama.cpp`:
 1. Define a new `llm_arch`
 2. Define the tensors layout in `LLM_TENSOR_NAMES`
-3. Add any non standard metadata in `llm_load_hparams`
+3. Add any non-standard metadata in `llm_load_hparams`
 4. Create the tensors for inference in `llm_load_tensors`
 5. If the model has a RoPE operation, add the rope type in `llama_rope_type`
 
@@ -98,7 +98,7 @@ This is the funniest part, you have to provide the inference graph implementatio
 
 Have a look at existing implementation like `build_llama`, `build_dbrx` or `build_bert`.
 
-When implementing a new graph, please note that the underlying `ggml` backends might not support them all, support for missing backend operations can be added in another PR.
+Some `ggml` backends do not support all operations, backend implementation can be added in a separate PR.
 
 Note: to debug the inference graph: you can use [llama-eval-callback](/examples/eval-callback/).
 

gguf-py/gguf/constants.py

@@ -242,6 +242,7 @@ class MODEL_ARCH(IntEnum):
     QWEN2VL          = auto()
     PHI2             = auto()
     PHI3             = auto()
+    PHIMOE           = auto()
     PLAMO            = auto()
     CODESHELL        = auto()
     ORION            = auto()
@@ -424,6 +425,7 @@ class MODEL_TENSOR(IntEnum):
     MODEL_ARCH.QWEN2VL:          "qwen2vl",
     MODEL_ARCH.PHI2:             "phi2",
     MODEL_ARCH.PHI3:             "phi3",
+    MODEL_ARCH.PHIMOE:           "phimoe",
     MODEL_ARCH.PLAMO:            "plamo",
     MODEL_ARCH.CODESHELL:        "codeshell",
     MODEL_ARCH.ORION:            "orion",
@@ -934,6 +936,24 @@ class MODEL_TENSOR(IntEnum):
         MODEL_TENSOR.FFN_DOWN,
         MODEL_TENSOR.FFN_UP,
     ],
+    MODEL_ARCH.PHIMOE: [
+        MODEL_TENSOR.TOKEN_EMBD,
+        MODEL_TENSOR.OUTPUT_NORM,
+        MODEL_TENSOR.OUTPUT,
+        MODEL_TENSOR.ROPE_FACTORS_LONG,
+        MODEL_TENSOR.ROPE_FACTORS_SHORT,
+        MODEL_TENSOR.ATTN_NORM,
+        MODEL_TENSOR.ATTN_QKV,
+        MODEL_TENSOR.ATTN_Q,
+        MODEL_TENSOR.ATTN_K,
+        MODEL_TENSOR.ATTN_V,
+        MODEL_TENSOR.ATTN_OUT,
+        MODEL_TENSOR.FFN_NORM,
+        MODEL_TENSOR.FFN_GATE_INP,
+        MODEL_TENSOR.FFN_GATE_EXP,
+        MODEL_TENSOR.FFN_DOWN_EXP,
+        MODEL_TENSOR.FFN_UP_EXP,
+    ],
     MODEL_ARCH.CODESHELL: [
         MODEL_TENSOR.TOKEN_EMBD,
         MODEL_TENSOR.POS_EMBD,

gguf-py/gguf/tensor_mapping.py

@@ -55,7 +55,7 @@ class TensorNameMap:
         # Output
         MODEL_TENSOR.OUTPUT: (
             "embed_out",                 # gptneox
-            "lm_head",                   # gpt2 mpt falcon llama-hf baichuan qwen mamba dbrx jais nemotron exaone olmoe olmo2
+            "lm_head",                   # gpt2 mpt falcon llama-hf baichuan qwen mamba dbrx jais nemotron exaone olmoe olmo2 phimoe
             "output",                    # llama-pth bloom internlm2
             "word_embeddings_for_head",  # persimmon
             "lm_head.linear",            # phi2
@@ -68,7 +68,7 @@ class TensorNameMap:
         MODEL_TENSOR.OUTPUT_NORM: (
             "gpt_neox.final_layer_norm",               # gptneox
             "transformer.ln_f",                        # gpt2 gpt-j falcon jais exaone
-            "model.norm",                              # llama-hf baichuan internlm2 olmoe olmo2
+            "model.norm",                              # llama-hf baichuan internlm2 olmoe olmo2 phimoe
             "norm",                                    # llama-pth
             "transformer.norm_f",                      # mpt dbrx
             "ln_f",                                    # refact bloom qwen gpt2
@@ -108,7 +108,7 @@ class TensorNameMap:
             "transformer.h.{bid}.input_layernorm",                  # falcon7b
             "h.{bid}.input_layernorm",                              # bloom
             "transformer.h.{bid}.ln_mlp",                           # falcon40b
-            "model.layers.{bid}.input_layernorm",                   # llama-hf nemotron olmoe
+            "model.layers.{bid}.input_layernorm",                   # llama-hf nemotron olmoe phimoe
             "layers.{bid}.attention_norm",                          # llama-pth
             "language_model.encoder.layers.{bid}.input_layernorm",  # persimmon
             "model.layers.{bid}.ln1",                               # yi
@@ -152,7 +152,7 @@ class TensorNameMap:
 
         # Attention query
         MODEL_TENSOR.ATTN_Q: (
-            "model.layers.{bid}.self_attn.q_proj",                       # llama-hf nemotron olmoe olmo2
+            "model.layers.{bid}.self_attn.q_proj",                       # llama-hf nemotron olmoe olmo2 phimoe
             "model.layers.{bid}.self_attn.q_proj_no_perm",               # llama-custom
             "layers.{bid}.attention.wq",                                 # llama-pth
             "encoder.layer.{bid}.attention.self.query",                  # bert
@@ -165,7 +165,7 @@ class TensorNameMap:
 
         # Attention key
         MODEL_TENSOR.ATTN_K: (
-            "model.layers.{bid}.self_attn.k_proj",                     # llama-hf nemotron olmoe olmo2
+            "model.layers.{bid}.self_attn.k_proj",                     # llama-hf nemotron olmoe olmo2 phimoe
             "model.layers.{bid}.self_attn.k_proj_no_perm",             # llama-custom
             "layers.{bid}.attention.wk",                               # llama-pth
             "encoder.layer.{bid}.attention.self.key",                  # bert
@@ -179,7 +179,7 @@ class TensorNameMap:
 
         # Attention value
         MODEL_TENSOR.ATTN_V: (
-            "model.layers.{bid}.self_attn.v_proj",                       # llama-hf nemotron olmoe olmo2
+            "model.layers.{bid}.self_attn.v_proj",                       # llama-hf nemotron olmoe olmo2 phimoe
             "layers.{bid}.attention.wv",                                 # llama-pth
             "encoder.layer.{bid}.attention.self.value",                  # bert
             "transformer.h.{bid}.attn.v_proj",                           # gpt-j
@@ -197,7 +197,7 @@ class TensorNameMap:
             "transformer.blocks.{bid}.attn.out_proj",                       # mpt
             "transformer.h.{bid}.self_attention.dense",                     # falcon
             "h.{bid}.self_attention.dense",                                 # bloom
-            "model.layers.{bid}.self_attn.o_proj",                          # llama-hf nemotron olmoe olmo2
+            "model.layers.{bid}.self_attn.o_proj",                          # llama-hf nemotron olmoe olmo2 phimoe
             "model.layers.{bid}.self_attn.linear_attn",                     # deci
             "layers.{bid}.attention.wo",                                    # llama-pth
             "encoder.layer.{bid}.attention.output.dense",                   # bert
@@ -242,7 +242,7 @@ class TensorNameMap:
             "transformer.h.{bid}.ln_2",                                      # gpt2 refact qwen jais exaone
             "h.{bid}.post_attention_layernorm",                              # bloom
             "transformer.blocks.{bid}.norm_2",                               # mpt
-            "model.layers.{bid}.post_attention_layernorm",                   # llama-hf nemotron olmoe
+            "model.layers.{bid}.post_attention_layernorm",                   # llama-hf nemotron olmoe phimoe
             "layers.{bid}.ffn_norm",                                         # llama-pth
             "language_model.encoder.layers.{bid}.post_attention_layernorm",  # persimmon
             "model.layers.{bid}.ln2",                                        # yi
@@ -265,7 +265,7 @@ class TensorNameMap:
 
         MODEL_TENSOR.FFN_GATE_INP: (
             "layers.{bid}.feed_forward.gate",                   # mixtral
-            "model.layers.{bid}.block_sparse_moe.gate",         # mixtral
+            "model.layers.{bid}.block_sparse_moe.gate",         # mixtral phimoe
             "model.layers.{bid}.mlp.gate",                      # qwen2moe olmoe
             "transformer.decoder_layer.{bid}.router",           # Grok
             "transformer.blocks.{bid}.ffn.router.layer",        # dbrx
@@ -306,10 +306,11 @@ class TensorNameMap:
         ),
 
         MODEL_TENSOR.FFN_UP_EXP: (
-            "layers.{bid}.feed_forward.experts.w3",          # mixtral (merged)
-            "transformer.decoder_layer.{bid}.moe.linear_v",  # Grok (merged)
-            "transformer.blocks.{bid}.ffn.experts.mlp.v1",   # dbrx
-            "model.layers.{bid}.mlp.experts.up_proj",        # qwen2moe olmoe (merged)
+            "layers.{bid}.feed_forward.experts.w3",           # mixtral (merged)
+            "transformer.decoder_layer.{bid}.moe.linear_v",   # Grok (merged)
+            "transformer.blocks.{bid}.ffn.experts.mlp.v1",    # dbrx
+            "model.layers.{bid}.mlp.experts.up_proj",         # qwen2moe olmoe (merged)
+            "model.layers.{bid}.block_sparse_moe.experts.w3", # phimoe (merged)
         ),
 
         MODEL_TENSOR.FFN_UP_SHEXP: (
@@ -338,10 +339,11 @@ class TensorNameMap:
         ),
 
         MODEL_TENSOR.FFN_GATE_EXP: (
-            "layers.{bid}.feed_forward.experts.w1",         # mixtral (merged)
-            "transformer.decoder_layer.{bid}.moe.linear",   # Grok (merged)
-            "transformer.blocks.{bid}.ffn.experts.mlp.w1",  # dbrx
-            "model.layers.{bid}.mlp.experts.gate_proj",     # qwen2moe olmoe (merged)
+            "layers.{bid}.feed_forward.experts.w1",           # mixtral (merged)
+            "transformer.decoder_layer.{bid}.moe.linear",     # Grok (merged)
+            "transformer.blocks.{bid}.ffn.experts.mlp.w1",    # dbrx
+            "model.layers.{bid}.mlp.experts.gate_proj",       # qwen2moe olmoe (merged)
+            "model.layers.{bid}.block_sparse_moe.experts.w1", # phimoe (merged)
         ),
 
         MODEL_TENSOR.FFN_GATE_SHEXP: (
@@ -383,6 +385,7 @@ class TensorNameMap:
             "transformer.blocks.{bid}.ffn.experts.mlp.w2",       # dbrx
             "model.layers.{bid}.mlp.experts.down_proj",          # qwen2moe olmoe (merged)
             "model.layers.{bid}.block_sparse_moe.output_linear", # granitemoe
+            "model.layers.{bid}.block_sparse_moe.experts.w2",    # phimoe (merged)
         ),
 
         MODEL_TENSOR.FFN_DOWN_SHEXP: (