Merge branch 'main' into patch-1

Author: abetlen

.github/workflows/build-and-release.yaml

@@ -29,7 +29,7 @@ jobs:
           python -m pip install -e .[all]
 
       - name: Build wheels
-        uses: pypa/cibuildwheel@v2.19.2
+        uses: pypa/cibuildwheel@v2.20.0
         env:
           # disable repair
           CIBW_REPAIR_WHEEL_COMMAND: ""
@@ -56,7 +56,7 @@ jobs:
           platforms: linux/arm64
 
       - name: Build wheels
-        uses: pypa/cibuildwheel@v2.19.2
+        uses: pypa/cibuildwheel@v2.20.0
         env:
           CIBW_SKIP: "*musllinux* pp*"
           CIBW_REPAIR_WHEEL_COMMAND: ""

.github/workflows/build-wheels-metal.yaml

@@ -30,7 +30,7 @@ jobs:
           python -m pip install -e .[all]
 
       - name: Build wheels
-        uses: pypa/cibuildwheel@v2.19.2
+        uses: pypa/cibuildwheel@v2.20.0
         env:
           # disable repair
           CIBW_REPAIR_WHEEL_COMMAND: ""

.github/workflows/generate-index-from-release.yaml

@@ -1,9 +1,11 @@
 name: Wheels Index
 
 on:
-  # Trigger on any new release
-  release:
-    types: [published]
+  # Trigger on new release
+  workflow_run:
+    workflows: ["Release", "Build Wheels (CUDA)", "Build Wheels (Metal)"]
+    types:
+      - completed
 
   # Allows you to run this workflow manually from the Actions tab
   workflow_dispatch:

CHANGELOG.md

@@ -7,6 +7,18 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.2.87]
+
+- feat: Update llama.cpp to ggerganov/llama.cpp@be55695eff44784a141a863f273661a6bce63dfc
+- fix: Include all llama.cpp source files and subdirectories by @abetlen in 9cad5714ae6e7c250af8d0bbb179f631368c928b
+- feat(ci): Re-build wheel index automatically when releases are created by @abetlen in 198f47dc1bd202fd2b71b29e041a9f33fe40bfad
+
+## [0.2.86]
+
+- feat: Update llama.cpp to ggerganov/llama.cpp@398ede5efeb07b9adf9fbda7ea63f630d476a792
+- feat: Ported back new grammar changes from C++ to Python implementation by @ExtReMLapin in (#1637)
+- fix: llama_grammar_accept_token arg order by @tc-wolf in (#1649)
+
 ## [0.2.85]
 
 - feat: Update llama.cpp to ggerganov/llama.cpp@398ede5efeb07b9adf9fbda7ea63f630d476a792

llama_cpp/__init__.py

@@ -1,4 +1,4 @@
 from .llama_cpp import *
 from .llama import *
 
-__version__ = "0.2.85"
+__version__ = "0.2.87"

llama_cpp/llama.py

@@ -777,11 +777,12 @@ def generate(
                 else:
                     break
             if longest_prefix > 0:
-                if self.verbose:
-                    print("Llama.generate: prefix-match hit", file=sys.stderr)
                 reset = False
                 tokens = tokens[longest_prefix:]
                 self.n_tokens = longest_prefix
+                if self.verbose:
+                    print(f"Llama.generate: {longest_prefix} prefix-match hit, "
+                          f"remaining {len(tokens)} prompt tokens to eval", file=sys.stderr)                    
 
         # Reset the model state
         if reset:
@@ -2159,7 +2160,7 @@ def from_pretrained(
 
         files = [
             file["name"] if isinstance(file, dict) else file
-            for file in hffs.ls(repo_id)
+            for file in hffs.ls(repo_id, recursive=True)
         ]
 
         # split each file into repo_id, subfolder, filename

llama_cpp/llama_cpp.py

@@ -944,7 +944,7 @@ class llama_context_params(ctypes.Structure):
 #     int32_t nthread;                     // number of threads to use for quantizing, if <=0 will use std::thread::hardware_concurrency()
 #     enum llama_ftype ftype;              // quantize to this llama_ftype
 #     enum ggml_type output_tensor_type;   // output tensor type
-#     enum ggml_type token_embedding_type; // itoken embeddings tensor type
+#     enum ggml_type token_embedding_type; // token embeddings tensor type
 #     bool allow_requantize;               // allow quantizing non-f32/f16 tensors
 #     bool quantize_output_tensor;         // quantize output.weight
 #     bool only_copy;                      // only copy tensors - ftype, allow_requantize and quantize_output_tensor are ignored
@@ -960,7 +960,7 @@ class llama_model_quantize_params(ctypes.Structure):
         nthread (int): number of threads to use for quantizing, if <=0 will use std::thread::hardware_concurrency()
         ftype (int): quantize to this llama_ftype
         output_tensor_type (int): output tensor type
-        token_embedding_type (int): itoken embeddings tensor type
+        token_embedding_type (int): token embeddings tensor type
         allow_requantize (bool): allow quantizing non-f32/f16 tensors
         quantize_output_tensor (bool): quantize output.weight
         only_copy (bool): only copy tensors - ftype, allow_requantize and quantize_output_tensor are ignored
@@ -3002,7 +3002,7 @@ def llama_grammar_init(
     n_rules: Union[ctypes.c_size_t, int],
     start_rule_index: Union[ctypes.c_size_t, int],
     /,
-) -> llama_grammar_p:
+) -> Optional[llama_grammar_p]:
     """Initialize a grammar from a set of rules."""
     ...