llama-cpp-python 0.2.64__tar.gz → 0.2.65__tar.gz

llama_cpp_python-0.2.65/.git/FETCH_HEAD

@@ -0,0 +1 @@
+65edc906712eb205afaedc7d69f27e33be162228		'65edc906712eb205afaedc7d69f27e33be162228' of https://github.com/abetlen/llama-cpp-python

llama_cpp_python-0.2.65/.git/HEAD

@@ -0,0 +1 @@
+65edc906712eb205afaedc7d69f27e33be162228

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/.git/config

@@ -9,7 +9,7 @@
 [gc]
 	auto = 0
 [http "https://github.com/"]
-	extraheader = AUTHORIZATION: basic eC1hY2Nlc3MtdG9rZW46Z2hzX3BzZ0pQR0U0Q2dEREFveWJBMTRqUXpvcHVTRXg2NTF2OTBJQQ==
+	extraheader = AUTHORIZATION: basic eC1hY2Nlc3MtdG9rZW46Z2hzX1RkWU05UXN2QXJkejl6bWowd2tvdzhxSjBuVlMzeDRjcHgxSg==
 [submodule "vendor/llama.cpp"]
 	active = true
 	url = https://github.com/ggerganov/llama.cpp.git

llama_cpp_python-0.2.65/.git/logs/HEAD

@@ -0,0 +1 @@
+0000000000000000000000000000000000000000 65edc906712eb205afaedc7d69f27e33be162228 runner <runner@fv-az979-609.lnrek1131jeubcfwnjesvv4kld.dx.internal.cloudapp.net> 1714140760 +0000	checkout: moving from master to refs/tags/v0.2.65

llama_cpp_python-0.2.65/.git/modules/vendor/llama.cpp/FETCH_HEAD

@@ -0,0 +1 @@
+46e12c4692a37bdd31a0432fc5153d7d22bc7f72		'46e12c4692a37bdd31a0432fc5153d7d22bc7f72' of https://github.com/ggerganov/llama.cpp

llama_cpp_python-0.2.65/.git/modules/vendor/llama.cpp/HEAD

@@ -0,0 +1 @@
+46e12c4692a37bdd31a0432fc5153d7d22bc7f72

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/.git/modules/vendor/llama.cpp/config

@@ -16,7 +16,7 @@
 [gc]
 	auto = 0
 [http "https://github.com/"]
-	extraheader = AUTHORIZATION: basic eC1hY2Nlc3MtdG9rZW46Z2hzX3BzZ0pQR0U0Q2dEREFveWJBMTRqUXpvcHVTRXg2NTF2OTBJQQ==
+	extraheader = AUTHORIZATION: basic eC1hY2Nlc3MtdG9rZW46Z2hzX1RkWU05UXN2QXJkejl6bWowd2tvdzhxSjBuVlMzeDRjcHgxSg==
 [url "https://github.com/"]
 	insteadOf = git@github.com:
 	insteadOf = org-6826477@github.com:

llama_cpp_python-0.2.65/.git/modules/vendor/llama.cpp/logs/HEAD

@@ -0,0 +1,2 @@
+0000000000000000000000000000000000000000 4b1c3c98b442a4c84a788fff6323f6fa7e678a5b runner <runner@fv-az979-609.lnrek1131jeubcfwnjesvv4kld.dx.internal.cloudapp.net> 1714140761 +0000	clone: from https://github.com/ggerganov/llama.cpp.git
+4b1c3c98b442a4c84a788fff6323f6fa7e678a5b 46e12c4692a37bdd31a0432fc5153d7d22bc7f72 runner <runner@fv-az979-609.lnrek1131jeubcfwnjesvv4kld.dx.internal.cloudapp.net> 1714140763 +0000	checkout: moving from master to 46e12c4692a37bdd31a0432fc5153d7d22bc7f72

llama_cpp_python-0.2.65/.git/modules/vendor/llama.cpp/logs/refs/heads/master

@@ -0,0 +1 @@
+0000000000000000000000000000000000000000 4b1c3c98b442a4c84a788fff6323f6fa7e678a5b runner <runner@fv-az979-609.lnrek1131jeubcfwnjesvv4kld.dx.internal.cloudapp.net> 1714140761 +0000	clone: from https://github.com/ggerganov/llama.cpp.git

llama_cpp_python-0.2.65/.git/modules/vendor/llama.cpp/logs/refs/remotes/origin/HEAD

@@ -0,0 +1 @@
+0000000000000000000000000000000000000000 4b1c3c98b442a4c84a788fff6323f6fa7e678a5b runner <runner@fv-az979-609.lnrek1131jeubcfwnjesvv4kld.dx.internal.cloudapp.net> 1714140761 +0000	clone: from https://github.com/ggerganov/llama.cpp.git

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/.git/modules/vendor/llama.cpp/modules/kompute/config

@@ -13,7 +13,7 @@
 [gc]
 	auto = 0
 [http "https://github.com/"]
-	extraheader = AUTHORIZATION: basic eC1hY2Nlc3MtdG9rZW46Z2hzX3BzZ0pQR0U0Q2dEREFveWJBMTRqUXpvcHVTRXg2NTF2OTBJQQ==
+	extraheader = AUTHORIZATION: basic eC1hY2Nlc3MtdG9rZW46Z2hzX1RkWU05UXN2QXJkejl6bWowd2tvdzhxSjBuVlMzeDRjcHgxSg==
 [url "https://github.com/"]
 	insteadOf = git@github.com:
 	insteadOf = org-6826477@github.com:

llama_cpp_python-0.2.65/.git/modules/vendor/llama.cpp/modules/kompute/logs/HEAD

@@ -0,0 +1,2 @@
+0000000000000000000000000000000000000000 d1e3b0953cf66acc94b2e29693e221427b2c1f3f runner <runner@fv-az979-609.lnrek1131jeubcfwnjesvv4kld.dx.internal.cloudapp.net> 1714140764 +0000	clone: from https://github.com/nomic-ai/kompute.git
+d1e3b0953cf66acc94b2e29693e221427b2c1f3f 4565194ed7c32d1d2efa32ceab4d3c6cae006306 runner <runner@fv-az979-609.lnrek1131jeubcfwnjesvv4kld.dx.internal.cloudapp.net> 1714140766 +0000	checkout: moving from master to 4565194ed7c32d1d2efa32ceab4d3c6cae006306

llama_cpp_python-0.2.65/.git/modules/vendor/llama.cpp/modules/kompute/logs/refs/heads/master

@@ -0,0 +1 @@
+0000000000000000000000000000000000000000 d1e3b0953cf66acc94b2e29693e221427b2c1f3f runner <runner@fv-az979-609.lnrek1131jeubcfwnjesvv4kld.dx.internal.cloudapp.net> 1714140764 +0000	clone: from https://github.com/nomic-ai/kompute.git

llama_cpp_python-0.2.65/.git/modules/vendor/llama.cpp/modules/kompute/logs/refs/remotes/origin/HEAD

@@ -0,0 +1 @@
+0000000000000000000000000000000000000000 d1e3b0953cf66acc94b2e29693e221427b2c1f3f runner <runner@fv-az979-609.lnrek1131jeubcfwnjesvv4kld.dx.internal.cloudapp.net> 1714140764 +0000	clone: from https://github.com/nomic-ai/kompute.git

llama_cpp_python-0.2.65/.git/modules/vendor/llama.cpp/objects/da/23b209c4dd5b2b9627f70aa12d22a8ae9772ac

@@ -0,0 +1,2 @@
+x
m��
+�@�=�)B{Qś���O�.d���A|w-(E�@���p2؜�M����X�:���^rq&%>ޮ�yT��<�i������	5�`�Gf�L*w�6H9�|l�\����/�����ĥڝa���W�

llama_cpp_python-0.2.65/.git/modules/vendor/llama.cpp/packed-refs

@@ -0,0 +1,2 @@
+# pack-refs with: peeled fully-peeled sorted 
+4b1c3c98b442a4c84a788fff6323f6fa7e678a5b refs/remotes/origin/master

llama_cpp_python-0.2.65/.git/modules/vendor/llama.cpp/refs/heads/master

@@ -0,0 +1 @@
+4b1c3c98b442a4c84a788fff6323f6fa7e678a5b

llama_cpp_python-0.2.65/.git/modules/vendor/llama.cpp/shallow

@@ -0,0 +1,2 @@
+46e12c4692a37bdd31a0432fc5153d7d22bc7f72
+4b1c3c98b442a4c84a788fff6323f6fa7e678a5b

llama_cpp_python-0.2.65/.git/refs/tags/v0.2.65

@@ -0,0 +1 @@
+65edc906712eb205afaedc7d69f27e33be162228

llama_cpp_python-0.2.65/.git/shallow

@@ -0,0 +1 @@
+65edc906712eb205afaedc7d69f27e33be162228

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/.github/workflows/test.yaml

@@ -57,7 +57,7 @@ jobs:
 
   build-macos:
 
-    runs-on: macos-latest
+    runs-on: macos-13
     strategy:
       matrix:
         python-version: ["3.8", "3.9", "3.10", "3.11", "3.12"]
@@ -107,7 +107,7 @@ jobs:
 
   build-macos-metal:
 
-    runs-on: macos-latest
+    runs-on: macos-13
 
     steps:
       - uses: actions/checkout@v3

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/CHANGELOG.md

@@ -7,6 +7,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.2.65]
+
+- feat: Update llama.cpp to ggerganov/llama.cpp@46e12c4692a37bdd31a0432fc5153d7d22bc7f72
+- feat: Allow for possibly non-pooled embeddings by @iamlemec in #1380
+
 ## [0.2.64]
 
 - feat: Update llama.cpp to ggerganov/llama.cpp@4e96a812b3ce7322a29a3008db2ed73d9087b176

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: llama_cpp_python
-Version: 0.2.64
+Version: 0.2.65
 Summary: Python bindings for the llama.cpp library
 Author-Email: Andrei Betlen <abetlen@gmail.com>
 License: MIT
@@ -619,7 +619,7 @@ llama = Llama(
 
 ### Embeddings
 
-To generate text embeddings use [`create_embedding`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama.create_embedding).
+To generate text embeddings use [`create_embedding`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama.create_embedding) or [`embed`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama.embed). Note that you must pass `embedding=True` to the constructor upon model creation for these to work properly.
 
 ```python
 import llama_cpp
@@ -633,6 +633,12 @@ embeddings = llm.create_embedding("Hello, world!")
 embeddings = llm.create_embedding(["Hello, world!", "Goodbye, world!"])
 ```
 
+There are two primary notions of embeddings in a Transformer-style model: *token level* and *sequence level*. Sequence level embeddings are produced by "pooling" token level embeddings together, usually by averaging them or using the first token.
+
+Models that are explicitly geared towards embeddings will usually return sequence level embeddings by default, one for each input string. Non-embedding models such as those designed for text generation will typically return only token level embeddings, one for each token in each sequence. Thus the dimensionality of the return type will be one higher for token level embeddings.
+
+It is possible to control pooling behavior in some cases using the `pooling_type` flag on model creation. You can ensure token level embeddings from any model using `LLAMA_POOLING_TYPE_NONE`. The reverse, getting a generation oriented model to yield sequence level embeddings is currently not possible, but you can always do the pooling manually.
+
 ### Adjusting the Context Window
 
 The context window of the Llama models determines the maximum number of tokens that can be processed at once. By default, this is set to 512 tokens, but can be adjusted based on your requirements.

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/README.md

@@ -575,7 +575,7 @@ llama = Llama(
 
 ### Embeddings
 
-To generate text embeddings use [`create_embedding`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama.create_embedding).
+To generate text embeddings use [`create_embedding`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama.create_embedding) or [`embed`](https://llama-cpp-python.readthedocs.io/en/latest/api-reference/#llama_cpp.Llama.embed). Note that you must pass `embedding=True` to the constructor upon model creation for these to work properly.
 
 ```python
 import llama_cpp
@@ -589,6 +589,12 @@ embeddings = llm.create_embedding("Hello, world!")
 embeddings = llm.create_embedding(["Hello, world!", "Goodbye, world!"])
 ```
 
+There are two primary notions of embeddings in a Transformer-style model: *token level* and *sequence level*. Sequence level embeddings are produced by "pooling" token level embeddings together, usually by averaging them or using the first token.
+
+Models that are explicitly geared towards embeddings will usually return sequence level embeddings by default, one for each input string. Non-embedding models such as those designed for text generation will typically return only token level embeddings, one for each token in each sequence. Thus the dimensionality of the return type will be one higher for token level embeddings.
+
+It is possible to control pooling behavior in some cases using the `pooling_type` flag on model creation. You can ensure token level embeddings from any model using `LLAMA_POOLING_TYPE_NONE`. The reverse, getting a generation oriented model to yield sequence level embeddings is currently not possible, but you can always do the pooling manually.
+
 ### Adjusting the Context Window
 
 The context window of the Llama models determines the maximum number of tokens that can be processed at once. By default, this is set to 512 tokens, but can be adjusted based on your requirements.

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/llama_cpp/__init__.py

@@ -1,4 +1,4 @@
 from .llama_cpp import *
 from .llama import *
 
-__version__ = "0.2.64"
+__version__ = "0.2.65"

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/llama_cpp/_internals.py

@@ -273,6 +273,10 @@ class _LlamaContext:
         assert self.ctx is not None
         return llama_cpp.llama_n_ctx(self.ctx)
 
+    def pooling_type(self) -> int:
+        assert self.ctx is not None
+        return llama_cpp.llama_pooling_type(self.ctx)
+
     def kv_cache_clear(self):
         assert self.ctx is not None
         llama_cpp.llama_kv_cache_clear(self.ctx)
@@ -641,6 +645,16 @@ def _should_add_bos(model: _LlamaModel) -> bool:
         return llama_cpp.llama_vocab_type(model.model) == llama_cpp.LLAMA_VOCAB_TYPE_SPM
 
 
+# Embedding functions
+
+
+def _normalize_embedding(embedding):
+    norm = float(np.linalg.norm(embedding))
+    if norm == 0.0:
+        return embedding
+    return [v / norm for v in embedding]
+
+
 # Python wrappers over common/sampling structs
 
 

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/llama_cpp/llama.py

@@ -50,6 +50,7 @@ from ._internals import (
     _LlamaTokenDataArray,  # type: ignore
     _LlamaSamplingParams,  # type: ignore
     _LlamaSamplingContext,  # type: ignore
+    _normalize_embedding,  # type: ignore
 )
 from ._logger import set_verbose
 from ._utils import suppress_stdout_stderr
@@ -760,7 +761,7 @@ class Llama:
         input = input if isinstance(input, list) else [input]
 
         # get numeric embeddings
-        embeds: List[List[float]]
+        embeds: Union[List[List[float]], List[List[List[float]]]]
         total_tokens: int
         embeds, total_tokens = self.embed(input, return_count=True)  # type: ignore
 
@@ -787,7 +788,7 @@ class Llama:
     def embed(
         self,
         input: Union[str, List[str]],
-        normalize: bool = True,
+        normalize: bool = False,
         truncate: bool = True,
         return_count: bool = False,
     ):
@@ -803,6 +804,10 @@ class Llama:
         n_embd = self.n_embd()
         n_batch = self.n_batch
 
+        # get pooling information
+        pooling_type = self.pooling_type()
+        logits_all = pooling_type == llama_cpp.LLAMA_POOLING_TYPE_NONE
+
         if self.context_params.embeddings == False:
             raise RuntimeError(
                 "Llama model must be created with embedding=True to call this method"
@@ -820,29 +825,37 @@ class Llama:
         self._batch.reset()
 
         # decode and fetch embeddings
-        data: List[List[float]] = []
+        data: Union[List[List[float]], List[List[List[float]]]] = []
 
-        def decode_batch(n_seq: int):
+        def decode_batch(seq_sizes: List[int]):
             assert self._ctx.ctx is not None
             llama_cpp.llama_kv_cache_clear(self._ctx.ctx)
             self._ctx.decode(self._batch)
             self._batch.reset()
 
             # store embeddings
-            for i in range(n_seq):
-                ptr = llama_cpp.llama_get_embeddings_seq(
-                    self._ctx.ctx, i
-                )
-                if not ptr:
-                    raise RuntimeError("Failed to get embeddings from sequence pooling type is not set")
-                embedding: List[float] = ptr[:n_embd]
-                if normalize:
-                    norm = float(np.linalg.norm(embedding))
-                    embedding = [v / norm for v in embedding]
-                data.append(embedding)
+            if pooling_type == llama_cpp.LLAMA_POOLING_TYPE_NONE:
+                pos: int = 0
+                for i, size in enumerate(seq_sizes):
+                    ptr = llama_cpp.llama_get_embeddings(self._ctx.ctx)
+                    embedding: List[List[float]] = [
+                        ptr[pos + j * n_embd : pos + (j + 1) * n_embd] for j in range(size)
+                    ]
+                    if normalize:
+                        embedding = [_normalize_embedding(e) for e in embedding]
+                    data.append(embedding)
+                    pos += size
+            else:
+                for i in range(len(seq_sizes)):
+                    ptr = llama_cpp.llama_get_embeddings_seq(self._ctx.ctx, i)
+                    embedding: List[float] = ptr[:n_embd]
+                    if normalize:
+                        embedding = _normalize_embedding(embedding)
+                    data.append(embedding)
 
         # init state
         total_tokens = 0
+        s_batch = []
         t_batch = 0
         p_batch = 0
 
@@ -863,17 +876,21 @@ class Llama:
 
             # time to eval batch
             if t_batch + n_tokens > n_batch:
-                decode_batch(p_batch)
+                decode_batch(s_batch)
+                s_batch = []
                 t_batch = 0
                 p_batch = 0
 
             # add to batch
-            self._batch.add_sequence(tokens, p_batch, False)
+            self._batch.add_sequence(tokens, p_batch, logits_all)
+
+            # update batch stats
+            s_batch.append(n_tokens)
             t_batch += n_tokens
             p_batch += 1
 
         # hanlde last batch
-        decode_batch(p_batch)
+        decode_batch(s_batch)
 
         if self.verbose:
             llama_cpp.llama_print_timings(self._ctx.ctx)
@@ -1845,6 +1862,10 @@ class Llama:
         """Return the newline token."""
         return self._model.token_nl()
 
+    def pooling_type(self) -> str:
+        """Return the pooling type."""
+        return self._ctx.pooling_type()
+
     @staticmethod
     def logits_to_logprobs(
         logits: Union[npt.NDArray[np.single], List], axis: int = -1

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/llama_cpp/llama_cpp.py

@@ -811,6 +811,7 @@ It might not exist for progress report where '.' is output repeatedly."""
 #     bool quantize_output_tensor;         // quantize output.weight
 #     bool only_copy;                      // only copy tensors - ftype, allow_requantize and quantize_output_tensor are ignored
 #     bool pure;                           // quantize all tensors to the default type
+#     bool keep_split;                     // quantize to the same number of shards
 #     void * imatrix;                      // pointer to importance matrix data
 #     void * kv_overrides;                 // pointer to vector containing overrides
 # } llama_model_quantize_params;
@@ -826,6 +827,7 @@ class llama_model_quantize_params(ctypes.Structure):
         quantize_output_tensor (bool): quantize output.weight
         only_copy (bool): only copy tensors - ftype, allow_requantize and quantize_output_tensor are ignored
         pure (bool): quantize all tensors to the default type
+        keep_split (bool): quantize to the same number of shards
         imatrix (ctypes.c_void_p): pointer to importance matrix data
         kv_overrides (ctypes.c_void_p): pointer to vector containing overrides
     """
@@ -839,6 +841,7 @@ class llama_model_quantize_params(ctypes.Structure):
         quantize_output_tensor: bool
         only_copy: bool
         pure: bool
+        keep_split: bool
         imatrix: ctypes.c_void_p
         kv_overrides: ctypes.c_void_p
 
@@ -851,6 +854,7 @@ class llama_model_quantize_params(ctypes.Structure):
         ("quantize_output_tensor", ctypes.c_bool),
         ("only_copy", ctypes.c_bool),
         ("pure", ctypes.c_bool),
+        ("keep_split", ctypes.c_bool),
         ("imatrix", ctypes.c_void_p),
         ("kv_overrides", ctypes.c_void_p),
     ]
@@ -1167,13 +1171,19 @@ def llama_n_seq_max(ctx: llama_context_p, /) -> int:
     ...
 
 
-# LLAMA_API enum llama_vocab_type llama_vocab_type(const struct llama_model * model);
+# LLAMA_API enum llama_pooling_type llama_pooling_type(const struct llama_context * ctx);
+@ctypes_function("llama_pooling_type", [llama_context_p_ctypes], ctypes.c_int)
+def llama_pooling_type(ctx: llama_context_p, /) -> int:
+    ...
+
+
+# LLAMA_API enum llama_vocab_type   llama_vocab_type  (const struct llama_model   * model);
 @ctypes_function("llama_vocab_type", [llama_model_p_ctypes], ctypes.c_int)
 def llama_vocab_type(model: llama_model_p, /) -> int:
     ...
 
 
-# LLAMA_API enum llama_rope_type  llama_rope_type (const struct llama_model * model);
+# LLAMA_API enum llama_rope_type    llama_rope_type   (const struct llama_model   * model);
 @ctypes_function("llama_rope_type", [llama_model_p_ctypes], ctypes.c_int)
 def llama_rope_type(model: llama_model_p, /) -> int:
     ...
@@ -3091,7 +3101,7 @@ def llama_sample_token_greedy(
     ...
 
 
-# /// @details Randomly selects a token from the candidates based on their probabilities.
+# /// @details Randomly selects a token from the candidates based on their probabilities using the RNG of ctx.
 # LLAMA_API llama_token llama_sample_token(
 #         struct llama_context * ctx,
 #       llama_token_data_array * candidates);

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/llama_cpp/llama_types.py

@@ -24,7 +24,7 @@ class EmbeddingUsage(TypedDict):
 class Embedding(TypedDict):
     index: int
     object: str
-    embedding: List[float]
+    embedding: Union[List[float], List[List[float]]]
 
 
 class CreateEmbeddingResponse(TypedDict):

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/llama_cpp/server/settings.py

@@ -2,8 +2,10 @@ from __future__ import annotations
 
 import multiprocessing
 
-from typing import Optional, List, Literal, Union
-from pydantic import Field, root_validator
+from typing import Optional, List, Literal, Union, Dict, cast
+from typing_extensions import Self
+
+from pydantic import Field, model_validator
 from pydantic_settings import BaseSettings
 
 import llama_cpp
@@ -173,15 +175,16 @@ class ModelSettings(BaseSettings):
         default=True, description="Whether to print debug information."
     )
 
-    @root_validator(pre=True)  # pre=True to ensure this runs before any other validation
-    def set_dynamic_defaults(cls, values):
+    @model_validator(mode="before")  # pre=True to ensure this runs before any other validation
+    def set_dynamic_defaults(self) -> Self:
         # If n_threads or n_threads_batch is -1, set it to multiprocessing.cpu_count()
         cpu_count = multiprocessing.cpu_count()
+        values = cast(Dict[str, int], self)
         if values.get('n_threads', 0) == -1:
             values['n_threads'] = cpu_count
         if values.get('n_threads_batch', 0) == -1:
             values['n_threads_batch'] = cpu_count
-        return values
+        return self
 
 
 class ServerSettings(BaseSettings):

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/vendor/llama.cpp/CMakeLists.txt

@@ -43,11 +43,7 @@ else()
     set(LLAMA_METAL_DEFAULT OFF)
 endif()
 
-if (CMAKE_SYSTEM_NAME MATCHES "ANDROID")
-    set(LLAMA_LLAMAFILE_DEFAULT OFF)
-else()
-    set(LLAMA_LLAMAFILE_DEFAULT ON)
-endif()
+set(LLAMA_LLAMAFILE_DEFAULT ON)
 
 # general
 option(BUILD_SHARED_LIBS                "build shared libraries"                                OFF)

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/vendor/llama.cpp/README.md

@@ -93,6 +93,7 @@ Typically finetunes of the base models below are supported as well.
 
 - [X] LLaMA 🦙
 - [x] LLaMA 2 🦙🦙
+- [x] LLaMA 3 🦙🦙🦙
 - [X] [Mistral 7B](https://huggingface.co/mistralai/Mistral-7B-v0.1)
 - [x] [Mixtral MoE](https://huggingface.co/models?search=mistral-ai/Mixtral)
 - [x] [DBRX](https://huggingface.co/databricks/dbrx-instruct)
@@ -119,8 +120,9 @@ Typically finetunes of the base models below are supported as well.
 - [x] [CodeShell](https://github.com/WisdomShell/codeshell)
 - [x] [Gemma](https://ai.google.dev/gemma)
 - [x] [Mamba](https://github.com/state-spaces/mamba)
+- [x] [Grok-1](https://huggingface.co/keyfan/grok-1-hf)
 - [x] [Xverse](https://huggingface.co/models?search=xverse)
-- [x] [Command-R](https://huggingface.co/CohereForAI/c4ai-command-r-v01)
+- [x] [Command-R models](https://huggingface.co/models?search=CohereForAI/c4ai-command-r)
 - [x] [SEA-LION](https://huggingface.co/models?search=sea-lion)
 - [x] [GritLM-7B](https://huggingface.co/GritLM/GritLM-7B) + [GritLM-8x7B](https://huggingface.co/GritLM/GritLM-8x7B)
 - [x] [OLMo](https://allenai.org/olmo)
@@ -135,6 +137,8 @@ Typically finetunes of the base models below are supported as well.
 - [x] [ShareGPT4V](https://huggingface.co/models?search=Lin-Chen/ShareGPT4V)
 - [x] [MobileVLM 1.7B/3B models](https://huggingface.co/models?search=mobileVLM)
 - [x] [Yi-VL](https://huggingface.co/models?search=Yi-VL)
+- [x] [Mini CPM](https://huggingface.co/models?search=MiniCPM)
+- [x] [Moondream](https://huggingface.co/vikhyatk/moondream2)
 
 **HTTP server**
 
@@ -1117,7 +1121,9 @@ docker run --gpus all -v /path/to/models:/models local/llama.cpp:server-cuda -m
 - Clean-up any trailing whitespaces, use 4 spaces for indentation, brackets on the same line, `void * ptr`, `int & a`
 - See [good first issues](https://github.com/ggerganov/llama.cpp/issues?q=is%3Aissue+is%3Aopen+label%3A%22good+first+issue%22) for tasks suitable for first contributions
 - Tensors store data in row-major order. We refer to dimension 0 as columns, 1 as rows, 2 as matrices
-- Matrix multiplication is unconventional: [`z = ggml_mul_mat(ctx, x, y)`](https://github.com/ggerganov/llama.cpp/blob/880e352277fc017df4d5794f0c21c44e1eae2b84/ggml.h#L1058-L1064) means `zT = x @ yT`
+- Matrix multiplication is unconventional: [`C = ggml_mul_mat(ctx, A, B)`](https://github.com/ggerganov/llama.cpp/blob/880e352277fc017df4d5794f0c21c44e1eae2b84/ggml.h#L1058-L1064) means $C^T = A B^T \Leftrightarrow C = B A^T.$
+
+![matmul](media/matmul.png)
 
 ### Docs
 

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/vendor/llama.cpp/ci/run.sh

@@ -160,7 +160,9 @@ function gg_run_test_scripts_debug {
 
     set -e
 
+    # TODO: too slow, run on dedicated node
     (cd ./examples/gguf-split && time bash tests.sh "$SRC/build-ci-debug/bin" "$MNT/models") 2>&1 | tee -a $OUT/${ci}-scripts.log
+   #(cd ./examples/quantize   && time bash tests.sh "$SRC/build-ci-debug/bin" "$MNT/models") 2>&1 | tee -a $OUT/${ci}-scripts.log
 
     set +e
 }
@@ -184,6 +186,7 @@ function gg_run_test_scripts_release {
     set -e
 
     (cd ./examples/gguf-split && time bash tests.sh "$SRC/build-ci-release/bin" "$MNT/models") 2>&1 | tee -a $OUT/${ci}-scripts.log
+    (cd ./examples/quantize   && time bash tests.sh "$SRC/build-ci-release/bin" "$MNT/models") 2>&1 | tee -a $OUT/${ci}-scripts.log
 
     set +e
 }

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/vendor/llama.cpp/common/build-info.cpp

@@ -1,4 +1,4 @@
 int LLAMA_BUILD_NUMBER = 1;
-char const *LLAMA_COMMIT = "4e96a81";
+char const *LLAMA_COMMIT = "46e12c4";
 char const *LLAMA_COMPILER = "cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0";
 char const *LLAMA_BUILD_TARGET = "x86_64-linux-gnu";

{llama_cpp_python-0.2.64 → llama_cpp_python-0.2.65}/vendor/llama.cpp/common/common.cpp

@@ -242,7 +242,9 @@ bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_pa
             invalid_param = true;
             return true;
         }
+        // This is temporary, in the future the samplign state will be moved fully to llama_sampling_context.
         params.seed = std::stoul(argv[i]);
+        sparams.seed = std::stoul(argv[i]);
         return true;
     }
     if (arg == "-t" || arg == "--threads") {
@@ -2326,12 +2328,12 @@ std::vector<llama_token> llama_tokenize(
     return result;
 }
 
-std::string llama_token_to_piece(const struct llama_context * ctx, llama_token token) {
+std::string llama_token_to_piece(const struct llama_context * ctx, llama_token token, bool special) {
     std::vector<char> result(8, 0);
-    const int n_tokens = llama_token_to_piece(llama_get_model(ctx), token, result.data(), result.size(), true);
+    const int n_tokens = llama_token_to_piece(llama_get_model(ctx), token, result.data(), result.size(), special);
     if (n_tokens < 0) {
         result.resize(-n_tokens);
-        int check = llama_token_to_piece(llama_get_model(ctx), token, result.data(), result.size(), true);
+        int check = llama_token_to_piece(llama_get_model(ctx), token, result.data(), result.size(), special);
         GGML_ASSERT(check == -n_tokens);
     } else {
         result.resize(n_tokens);