Merge tag 'v0.2.16' into main

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

@@ -33,6 +33,9 @@ jobs:
 
       - name: Build wheels
         run: python -m cibuildwheel --output-dir wheelhouse
+        env:
+          # disable repair
+          CIBW_REPAIR_WHEEL_COMMAND: ""
 
       - uses: actions/upload-artifact@v3
         with:

CHANGELOG.md

@@ -7,6 +7,19 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.2.16]
+
+- Update llama.cpp to ggerganov/llama.cp@a75fa576abba9d37f463580c379e4bbf1e1ad03c
+- Add `set_seed` to `Llama` class by @abetlen in fd41ed3a908761d286102a019a34c2938a15118d
+- Fix server doc arguments by @kjunggithub in #892
+- Fix response_format handler in llava chat handler by @abetlen in b62c44983921197ed10a7d29dc4ba920e9979380
+- Fix default max_tokens, chat completion is now unlimited (to context length) and completion is 16 tokens to match OpenAI defaults by @abetlen in e7962d2c733cbbeec5a37392c81f64185a9a39e8
+- Fix json_schema_to_gbnf helper so that it takes a json schema string as input instead by @abetlen in faeae181b1e868643c0dc28fcf039f077baf0829
+- Add support for $ref and $def in json_schema_to_gbnf to handle more complex function schemas by @abetlen in 770df344369c0630df1be14be9f9e301e7c56d24
+- Update functionary chat handler for new OpenAI api by abetlen in 1b376c62b775b401653facf25a519d116aafe99a
+- Fix add default stop sequence to chatml chat format by @abetlen in b84d76a844149216d511cfd8cdb9827148a1853c
+- Fix sampling bug when logits_all=False by @abetlen in 6f0b0b1b840af846938ed74d0e8170a91c40e617
+
 ## [0.2.15]
 
 - Update llama.cpp to ggerganov/llama.cpp@0a7c980b6f94a049cb804573df2d8092a34df8e4

llama_cpp/__init__.py

@@ -1,4 +1,4 @@
 from .llama_cpp import *
 from .llama import *
 
-__version__ = "0.2.15"
+__version__ = "0.2.16"

llama_cpp/llama.py

@@ -1019,27 +1019,26 @@ class Llama:
         """
         assert self._ctx.ctx is not None
         assert self._batch.batch is not None
-        n_ctx = self._n_ctx
+        self._ctx.kv_cache_seq_rm(-1, self.n_tokens, -1)
         for i in range(0, len(tokens), self.n_batch):
             batch = tokens[i : min(len(tokens), i + self.n_batch)]
-            n_past = min(n_ctx - len(batch), self.n_tokens)
+            n_past = self.n_tokens
             n_tokens = len(batch)
-            self._ctx.kv_cache_seq_rm(-1, n_past, -1)
             self._batch.set_batch(
                 batch=batch, n_past=n_past, logits_all=self.context_params.logits_all
             )
             self._ctx.decode(self._batch)
             # Save tokens
-            self.input_ids[self.n_tokens : self.n_tokens + n_tokens] = batch
+            self.input_ids[n_past : n_past + n_tokens] = batch
             # Save logits
-            rows = n_tokens if self.context_params.logits_all else 1
+            rows = n_tokens
             cols = self._n_vocab
             offset = (
                 0 if self.context_params.logits_all else n_tokens - 1
             )  # NOTE: Only save the last token logits if logits_all is False
-            self.scores[self.n_tokens + offset : self.n_tokens + n_tokens, :].reshape(
+            self.scores[n_past + offset : n_past + n_tokens, :].reshape(
                 -1
-            )[:] = self._ctx.get_logits()[: rows * cols]
+            )[:] = self._ctx.get_logits()[offset * cols: rows * cols]
             # Update n_tokens
             self.n_tokens += n_tokens
 

tests/test_llama.py

@@ -1,4 +1,7 @@
+import ctypes
+
 import pytest
+
 import llama_cpp
 
 MODEL = "./vendor/llama.cpp/models/ggml-vocab-llama.gguf"
@@ -36,19 +39,20 @@ def test_llama_cpp_tokenization():
 
 
 def test_llama_patch(monkeypatch):
-    llama = llama_cpp.Llama(model_path=MODEL, vocab_only=True)
+    n_ctx = 128
+    llama = llama_cpp.Llama(model_path=MODEL, vocab_only=True, n_ctx=n_ctx)
     n_vocab = llama_cpp.llama_n_vocab(llama._model.model)
+    assert n_vocab == 32000
 
     ## Set up mock function
-    def mock_eval(*args, **kwargs):
+    def mock_decode(*args, **kwargs):
         return 0
 
     def mock_get_logits(*args, **kwargs):
-        return (llama_cpp.c_float * n_vocab)(
+        size = n_vocab * n_ctx
-            *[llama_cpp.c_float(0) for _ in range(n_vocab)]
+        return (llama_cpp.c_float * size)()
-        )
 
-    monkeypatch.setattr("llama_cpp.llama_cpp.llama_decode", mock_eval)
+    monkeypatch.setattr("llama_cpp.llama_cpp.llama_decode", mock_decode)
     monkeypatch.setattr("llama_cpp.llama_cpp.llama_get_logits", mock_get_logits)
 
     output_text = " jumps over the lazy dog."
@@ -126,19 +130,19 @@ def test_llama_pickle():
 
 
 def test_utf8(monkeypatch):
-    llama = llama_cpp.Llama(model_path=MODEL, vocab_only=True)
+    n_ctx = 512
+    llama = llama_cpp.Llama(model_path=MODEL, vocab_only=True, n_ctx=n_ctx, logits_all=True)
     n_vocab = llama.n_vocab()
 
     ## Set up mock function
-    def mock_eval(*args, **kwargs):
+    def mock_decode(*args, **kwargs):
         return 0
 
     def mock_get_logits(*args, **kwargs):
-        return (llama_cpp.c_float * n_vocab)(
+        size = n_vocab * n_ctx
-            *[llama_cpp.c_float(0) for _ in range(n_vocab)]
+        return (llama_cpp.c_float * size)()
-        )
 
-    monkeypatch.setattr("llama_cpp.llama_cpp.llama_decode", mock_eval)
+    monkeypatch.setattr("llama_cpp.llama_cpp.llama_decode", mock_decode)
     monkeypatch.setattr("llama_cpp.llama_cpp.llama_get_logits", mock_get_logits)
 
     output_text = "😀"

vendor/llama.cpp

@@ -1 +1 @@
-Subproject commit 875fb42871a0f5a88fbe31a0b5edd697b84038e4
+Subproject commit a75fa576abba9d37f463580c379e4bbf1e1ad03c