Merge branch 'main' of github.com:abetlen/llama-cpp-python into better-server-params-and-fields

.dockerignore

@@ -0,0 +1,166 @@
+_skbuild/
+
+.envrc
+
+models/
+
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/#use-with-ide
+.pdm.toml
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+.idea/

.github/workflows/build-docker.yaml

@@ -0,0 +1,39 @@
+name: Build Docker
+
+on: workflow_dispatch
+
+permissions:
+  contents: write
+  packages: write
+
+jobs:
+  docker:
+    name: Build and push Docker image
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+        with:
+          submodules: "true"
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v2
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v2
+
+      - name: Login to GitHub Container Registry
+        uses: docker/login-action@v2 
+        with:
+          registry: ghcr.io
+          username: ${{ github.repository_owner }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Build and push
+        uses: docker/build-push-action@v4
+        with:
+          context: .
+          push: true # push to registry
+          pull: true # always fetch the latest base images
+          platforms: linux/amd64,linux/arm64 # build for both amd64 and arm64
+          tags: ghcr.io/abetlen/llama-cpp-python:latest

Dockerfile

@@ -0,0 +1,15 @@
+FROM python:3-bullseye
+
+# We need to set the host to 0.0.0.0 to allow outside access
+ENV HOST 0.0.0.0
+
+COPY . .
+
+# Install the package
+RUN apt update && apt install -y libopenblas-dev
+RUN python -m pip install --upgrade pip pytest cmake scikit-build setuptools fastapi uvicorn sse-starlette
+
+RUN LLAMA_OPENBLAS=1 python3 setup.py develop
+
+# Run the server
+CMD python3 -m llama_cpp.server

Dockerfile.cuda

@@ -0,0 +1,15 @@
+FROM nvidia/cuda:12.1.1-devel-ubuntu20.04
+
+# We need to set the host to 0.0.0.0 to allow outside access
+ENV HOST 0.0.0.0
+
+COPY . .
+
+# Install the package
+RUN apt update && apt install -y python3 python3-pip
+RUN python3 -m pip install --upgrade pip pytest cmake scikit-build setuptools fastapi uvicorn sse-starlette
+
+RUN LLAMA_CUBLAS=1 python3 setup.py develop
+
+# Run the server
+CMD python3 -m llama_cpp.server

README.md

@@ -72,6 +72,14 @@ python3 -m llama_cpp.server
 
 Navigate to [http://localhost:8000/docs](http://localhost:8000/docs) to see the OpenAPI documentation.
 
+## Docker image
+
+A Docker image is available on [GHCR](https://ghcr.io/abetlen/llama-cpp-python). To run the server:
+
+```bash
+docker run --rm -it -p8000:8000 -v /path/to/models:/models -eMODEL=/models/ggml-model-name.bin ghcr.io/abetlen/llama-cpp-python:latest
+```
+
 ## Low-level API
 
 The low-level API is a direct `ctypes` binding to the C API provided by `llama.cpp`.

llama_cpp/llama.py

@@ -53,12 +53,14 @@ class LlamaState:
     def __init__(
         self,
         eval_tokens: Deque[llama_cpp.llama_token],
-        eval_logits: Deque[List[float]],
+        eval_logits: Deque[List[llama_cpp.c_float]],
         llama_state,
+        llama_state_size: llama_cpp.c_size_t,
     ):
         self.eval_tokens = eval_tokens
         self.eval_logits = eval_logits
         self.llama_state = llama_state
+        self.llama_state_size = llama_state_size
 
 
 class Llama:
@@ -394,7 +396,7 @@ class Llama:
             and tuple(self.eval_tokens) == tuple(tokens[: len(self.eval_tokens)])
         ):
             if self.verbose:
-                print("generate cache hit", file=sys.stderr)
+                print("Llama.generate: cache hit", file=sys.stderr)
             reset = False
             tokens = tokens[len(self.eval_tokens) :]
 
@@ -516,7 +518,7 @@ class Llama:
 
         if self.cache and prompt_tokens in self.cache:
             if self.verbose:
-                print("cache hit", file=sys.stderr)
+                print("Llama._create_completion: cache hit", file=sys.stderr)
             self.load_state(self.cache[prompt_tokens])
 
         finish_reason = "length"
@@ -536,7 +538,7 @@ class Llama:
             if self.cache and len(completion_tokens) == 0:
                 if prompt_tokens not in self.cache:
                     if self.verbose:
-                        print("cache miss", file=sys.stderr)
+                        print("Llama._create_completion: cache miss", file=sys.stderr)
                     self.cache[prompt_tokens] = self.save_state()
 
             completion_tokens.append(token)
@@ -950,19 +952,25 @@ class Llama:
         assert self.ctx is not None
         state_size = llama_cpp.llama_get_state_size(self.ctx)
         llama_state = (llama_cpp.c_uint8 * int(state_size))()
-        if llama_cpp.llama_copy_state_data(self.ctx, llama_state) != state_size:
+        n_bytes = llama_cpp.llama_copy_state_data(self.ctx, llama_state)
+        if int(n_bytes) > int(state_size):
             raise RuntimeError("Failed to copy llama state data")
+        llama_state_compact = (llama_cpp.c_uint8 * int(n_bytes))()
+        llama_cpp.ctypes.memmove(llama_state_compact, llama_state, int(n_bytes))
+        if self.verbose:
+            print(f"Llama.save_state: saving {n_bytes} bytes of llama state", file=sys.stderr)
         return LlamaState(
             eval_tokens=self.eval_tokens.copy(),
             eval_logits=self.eval_logits.copy(),
-            llama_state=llama_state,
+            llama_state=llama_state_compact,
+            llama_state_size=n_bytes,
         )
 
     def load_state(self, state: LlamaState) -> None:
         assert self.ctx is not None
         self.eval_tokens = state.eval_tokens.copy()
         self.eval_logits = state.eval_logits.copy()
-        state_size = llama_cpp.llama_get_state_size(self.ctx)
+        state_size = state.llama_state_size
         if llama_cpp.llama_set_state_data(self.ctx, state.llama_state) != state_size:
             raise RuntimeError("Failed to set llama state data")
 

llama_cpp/llama_cpp.py

@@ -71,7 +71,7 @@ LLAMA_FILE_VERSION = ctypes.c_int(1)
 LLAMA_FILE_MAGIC = b"ggjt"
 LLAMA_FILE_MAGIC_UNVERSIONED = b"ggml"
 LLAMA_SESSION_MAGIC = b"ggsn"
-LLAMA_SESSION_VERSION = ctypes.c_int(0)
+LLAMA_SESSION_VERSION = ctypes.c_int(1)
 
 llama_context_p = c_void_p
 
@@ -136,9 +136,9 @@ LLAMA_FTYPE_MOSTLY_Q4_1_SOME_F16 = ctypes.c_int(
 )  # tok_embeddings.weight and output.weight are F16
 LLAMA_FTYPE_MOSTLY_Q4_2 = ctypes.c_int(5)  # except 1d tensors
 # LLAMA_FTYPE_MOSTYL_Q4_3 = ctypes.c_int(6)  # except 1d tensors
-LLAMA_FTYPE_MOSTYL_Q8_0 = ctypes.c_int(7)  # except 1d tensors
-LLAMA_FTYPE_MOSTYL_Q5_0 = ctypes.c_int(8)  # except 1d tensors
-LLAMA_FTYPE_MOSTYL_Q5_1 = ctypes.c_int(9)  # except 1d tensors
+LLAMA_FTYPE_MOSTLY_Q8_0 = ctypes.c_int(7)  # except 1d tensors
+LLAMA_FTYPE_MOSTLY_Q5_0 = ctypes.c_int(8)  # except 1d tensors
+LLAMA_FTYPE_MOSTLY_Q5_1 = ctypes.c_int(9)  # except 1d tensors
 
 # Functions
 
@@ -239,7 +239,8 @@ _lib.llama_set_rng_seed.argtypes = [llama_context_p, c_int]
 _lib.llama_set_rng_seed.restype = None
 
 
-# Returns the size in bytes of the state (rng, logits, embedding and kv_cache)
+# Returns the maximum size in bytes of the state (rng, logits, embedding
+# and kv_cache) - will often be smaller after compacting tokens
 def llama_get_state_size(ctx: llama_context_p) -> c_size_t:
     return _lib.llama_get_state_size(ctx)
 

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "llama_cpp_python"
-version = "0.1.40"
+version = "0.1.41"
 description = "Python bindings for the llama.cpp library"
 authors = ["Andrei Betlen <abetlen@gmail.com>"]
 license = "MIT"

setup.py

@@ -10,7 +10,7 @@ setup(
     description="A Python wrapper for llama.cpp",
     long_description=long_description,
     long_description_content_type="text/markdown",
-    version="0.1.40",
+    version="0.1.41",
     author="Andrei Betlen",
     author_email="abetlen@gmail.com",
     license="MIT",

vendor/llama.cpp

@@ -1 +1 @@
-Subproject commit 58b367c2d757c0ea12aec672382462b42204c724
+Subproject commit e216aa04633892b972d013719e38b59fd4917341