convert : refactor rope_freqs generation (ggml-org#9396)

* convert : refactor rope_freqs generation

This should also fix vocab-only conversion for Phi-3.

* convert : adapt MiniCPM3 to separate rope_freqs insertion

MiniCPM3's tokenizer is treated as a SentencePiece tokenizer to avoid
having to run its custom Python code which mixes tokenization
in the same file as tool calls.

gguf-py : add long and short RoPE factors to tensor mappings

Empty, but the key names are used to populate the mappings.

Author: compilade

convert_hf_to_gguf.py

@@ -15,6 +15,7 @@
 from pathlib import Path
 from hashlib import sha256
 from typing import TYPE_CHECKING, Any, Callable, ContextManager, Iterable, Iterator, Literal, Sequence, TypeVar, cast
+from itertools import chain
 
 import math
 import numpy as np
@@ -64,15 +65,14 @@ class Model:
     model_name: str | None
     metadata_override: Path | None
     dir_model_card: Path
-    is_lora: bool
 
     # subclasses should define this!
     model_arch: gguf.MODEL_ARCH
 
     def __init__(self, dir_model: Path, ftype: gguf.LlamaFileType, fname_out: Path, is_big_endian: bool = False,
                  use_temp_file: bool = False, eager: bool = False,
                  metadata_override: Path | None = None, model_name: str | None = None,
-                 split_max_tensors: int = 0, split_max_size: int = 0, dry_run: bool = False, small_first_shard: bool = False, is_lora: bool = False):
+                 split_max_tensors: int = 0, split_max_size: int = 0, dry_run: bool = False, small_first_shard: bool = False):
         if type(self) is Model:
             raise TypeError(f"{type(self).__name__!r} should not be directly instantiated")
 
@@ -94,7 +94,6 @@ def __init__(self, dir_model: Path, ftype: gguf.LlamaFileType, fname_out: Path,
         self.metadata_override = metadata_override
         self.model_name = model_name
         self.dir_model_card = dir_model  # overridden in convert_lora_to_gguf.py
-        self.is_lora = is_lora  # true if model is used inside convert_lora_to_gguf.py
 
         # Apply heuristics to figure out typical tensor encoding based on first layer tensor encoding type
         if self.ftype == gguf.LlamaFileType.GUESSED:
@@ -270,10 +269,14 @@ def tensor_force_quant(self, name: str, new_name: str, bid: int | None, n_dims:
 
         return False
 
+    # some models need extra generated tensors (like rope_freqs)
+    def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
+        return ()
+
     def prepare_tensors(self):
         max_name_len = max(len(s) for _, s in self.tensor_map.mapping.values()) + len(".weight,")
 
-        for name, data_torch in self.get_tensors():
+        for name, data_torch in chain(self.generate_extra_tensors(), self.get_tensors()):
             # we don't need these
             if name.endswith((".attention.masked_bias", ".attention.bias", ".rotary_emb.inv_freq")):
                 continue
@@ -1617,7 +1620,7 @@ def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iter
 
         return [(self.map_tensor_name(name), data_torch)]
 
-    def prepare_tensors(self):
+    def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
         if rope_scaling := self.find_hparam(["rope_scaling"], optional=True):
             if rope_scaling.get("rope_type", '').lower() == "llama3":
                 base = self.hparams.get("rope_theta", 10000.0)
@@ -1644,9 +1647,9 @@ def prepare_tensors(self):
                         smooth = (old_context_len / wavelen - low_freq_factor) / (high_freq_factor - low_freq_factor)
                         rope_factors.append(1 / ((1 - smooth) / factor + smooth))
 
-                if not self.is_lora:
-                    self.gguf_writer.add_tensor(self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FREQS), np.array(rope_factors, dtype=np.float32))
+                yield (self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FREQS), torch.tensor(rope_factors, dtype=torch.float32))
 
+    def prepare_tensors(self):
         super().prepare_tensors()
 
         if self._experts is not None:
@@ -1870,8 +1873,6 @@ class MiniCPM3Model(Model):
     def set_gguf_parameters(self):
         hparams = self.hparams
 
-        rope_dims = hparams["qk_rope_head_dim"]
-
         self.gguf_writer.add_file_type(self.ftype)
         self.gguf_writer.add_context_length(hparams["max_position_embeddings"])
         self.gguf_writer.add_embedding_length(hparams["hidden_size"])
@@ -1887,24 +1888,25 @@ def set_gguf_parameters(self):
         self.gguf_writer.add_key_length(hparams["qk_nope_head_dim"] + hparams["qk_rope_head_dim"])
         self.gguf_writer.add_rope_dimension_count(hparams["qk_rope_head_dim"])
 
+    def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
         rope_scaling = self.find_hparam(['rope_scaling'], True)
-        if rope_scaling is None:
-            return
+        if rope_scaling is not None:
+            rope_dims = self.hparams["qk_rope_head_dim"]
 
-        long_factors = rope_scaling.get('long_factor', None)
-        short_factors = rope_scaling.get('short_factor', None)
+            long_factors = rope_scaling.get('long_factor', None)
+            short_factors = rope_scaling.get('short_factor', None)
 
-        if long_factors is None or short_factors is None:
-            raise KeyError('Missing the required key rope_scaling.long_factor or rope_scaling_short_factor')
+            if long_factors is None or short_factors is None:
+                raise KeyError('Missing the required key rope_scaling.long_factor or rope_scaling_short_factor')
 
-        if len(long_factors) != len(short_factors) or len(long_factors) != rope_dims / 2:
-            raise ValueError(f'The length of rope long and short factors must be {rope_dims / 2}')
+            if len(long_factors) != len(short_factors) or len(long_factors) != rope_dims / 2:
+                raise ValueError(f'The length of rope long and short factors must be {rope_dims / 2}')
 
-        self.gguf_writer.add_tensor(gguf.TENSOR_NAMES[gguf.MODEL_TENSOR.ROPE_FACTORS_LONG]  + ".weight", np.array(long_factors, dtype=np.float32))
-        self.gguf_writer.add_tensor(gguf.TENSOR_NAMES[gguf.MODEL_TENSOR.ROPE_FACTORS_SHORT] + ".weight", np.array(short_factors, dtype=np.float32))
+            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))
 
     def set_vocab(self):
-        self._set_vocab_llama_hf()
+        self._set_vocab_sentencepiece()
 
     def _reverse_hf_permute(self, weights: Tensor, n_head: int, n_kv_head: int | None = None) -> Tensor:
         if n_kv_head is not None and n_head != n_kv_head:
@@ -2216,6 +2218,13 @@ def set_gguf_parameters(self):
         self.gguf_writer.add_file_type(self.ftype)
         self.gguf_writer.add_sliding_window(self.find_hparam(["sliding_window"]))
 
+    def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
+        n_embd = self.find_hparam(["hidden_size", "n_embd"])
+        n_head = self.find_hparam(["num_attention_heads", "n_head"])
+        max_pos_embds = self.find_hparam(["n_positions", "max_position_embeddings"])
+        orig_max_pos_embds = self.find_hparam(["original_max_position_embeddings"])
+        rope_dims = n_embd // n_head
+
         # write rope scaling for long context (128k) model
         rope_scaling = self.find_hparam(['rope_scaling'], True)
         if rope_scaling is None:
@@ -2245,9 +2254,8 @@ def set_gguf_parameters(self):
         if len(long_factors) != len(short_factors) or len(long_factors) != rope_dims / 2:
             raise ValueError(f'The length of rope long and short factors must be {rope_dims / 2}')
 
-        if not self.is_lora:
-            self.gguf_writer.add_tensor(gguf.TENSOR_NAMES[gguf.MODEL_TENSOR.ROPE_FACTORS_LONG]  + ".weight", np.array(long_factors, dtype=np.float32))
-            self.gguf_writer.add_tensor(gguf.TENSOR_NAMES[gguf.MODEL_TENSOR.ROPE_FACTORS_SHORT] + ".weight", np.array(short_factors, dtype=np.float32))
+        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("PlamoForCausalLM")
@@ -4071,7 +4079,7 @@ def set_gguf_parameters(self):
                 self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.LINEAR)
                 self.gguf_writer.add_rope_scaling_factor(hparams["rope_scaling"]["factor"])
 
-    def prepare_tensors(self):
+    def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
         if rope_scaling := self.find_hparam(["rope_scaling"], optional=True):
             if rope_scaling.get("rope_type", '').lower() == "llama3":
                 base = self.hparams.get("rope_theta", 10000.0)
@@ -4098,10 +4106,7 @@ def prepare_tensors(self):
                         smooth = (old_context_len / wavelen - low_freq_factor) / (high_freq_factor - low_freq_factor)
                         rope_factors.append(1 / ((1 - smooth) / factor + smooth))
 
-                if not self.is_lora:
-                    self.gguf_writer.add_tensor(self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FREQS), np.array(rope_factors, dtype=np.float32))
-
-        super().prepare_tensors()
+                yield (self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FREQS), torch.tensor(rope_factors, dtype=torch.float32))
 
 
 @Model.register("GraniteForCausalLM")

convert_lora_to_gguf.py

@@ -331,6 +331,10 @@ def set_gguf_parameters(self):
                 self.gguf_writer.add_float32(gguf.Keys.Adapter.LORA_ALPHA, self.lora_alpha)
                 super().set_gguf_parameters()
 
+            def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
+                # Never add extra tensors (e.g. rope_freqs) for LoRA adapters
+                return ()
+
             def get_tensors(self) -> Iterator[tuple[str, Tensor]]:
                 tensor_map: dict[str, PartialLoraTensor] = {}
 
@@ -392,7 +396,6 @@ def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iter
             dry_run=args.dry_run,
             dir_lora_model=dir_lora,
             lora_alpha=alpha,
-            is_lora=True,
         )
 
         logger.info("Exporting model...")

gguf-py/gguf/constants.py

@@ -814,6 +814,8 @@ class MODEL_TENSOR(IntEnum):
         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,
@@ -892,6 +894,8 @@ class MODEL_TENSOR(IntEnum):
         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_Q_A,
         MODEL_TENSOR.ATTN_Q_B,

gguf-py/gguf/tensor_mapping.py

@@ -87,6 +87,9 @@ class TensorNameMap:
             "rope.freqs",  # llama-pth
             "rotary_pos_emb.inv_freq",  # chatglm
         ),
+
+        MODEL_TENSOR.ROPE_FACTORS_LONG: (),
+        MODEL_TENSOR.ROPE_FACTORS_SHORT: (),
     }
 
     block_mappings_cfg: dict[MODEL_TENSOR, tuple[str, ...]] = {