Skip to content

Commit f8fcb3e

Browse files
abetlenabetlen
authored
feat: Update sampling API for llama.cpp (#1742)
* Initial samplng api update * Fix logger * Update tests * Update * Remove seed * Add sampling chain * Remove unnused test * Use Qwen2 0.5B for ci tests * Fix typo * Fix typo * Update cache version * Use real model for tests * Add huggingface-hub as a test dependency * Remove RUST_LOG=trace * Add actual logit processor test
1 parent a4e1451 commit f8fcb3e

File tree

10 files changed

+1035
-2083
lines changed

10 files changed

+1035
-2083
lines changed

‎.github/workflows/test.yaml‎

Lines changed: 54 additions & 27 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,4 @@
11
name: Tests
2-
32
on:
43
pull_request:
54
branches:
@@ -8,14 +7,34 @@ on:
87
branches:
98
- main
109

10+
env:
11+
REPO_ID: Qwen/Qwen2-0.5B-Instruct-GGUF
12+
MODEL_FILE: qwen2-0_5b-instruct-q8_0.gguf
13+
1114
jobs:
12-
build-linux:
15+
download-model:
16+
runs-on: ubuntu-latest
17+
steps:
18+
- name: Set up Python
19+
uses: actions/setup-python@v5
20+
with:
21+
python-version: "3.9"
22+
- name: Install huggingface-hub
23+
run: pip install huggingface-hub
24+
- name: Download model
25+
run: huggingface-cli download ${{ env.REPO_ID }} ${{ env.MODEL_FILE }}
26+
- name: Cache model
27+
uses: actions/cache@v4
28+
with:
29+
path: ~/.cache/huggingface/hub
30+
key: ${{ runner.os }}-model-${{ env.REPO_ID }}-${{ env.MODEL_FILE }}
1331

32+
build-linux:
33+
needs: download-model
1434
runs-on: ubuntu-latest
1535
strategy:
1636
matrix:
1737
python-version: ["3.9", "3.10", "3.11", "3.12"]
18-
1938
steps:
2039
- uses: actions/checkout@v4
2140
with:
@@ -26,36 +45,35 @@ jobs:
2645
with:
2746
python-version: ${{ matrix.python-version }}
2847
cache: 'pip'
29-
48+
- name: Restore model cache
49+
uses: actions/cache@v3
50+
with:
51+
path: ~/.cache/huggingface/hub
52+
key: ${{ runner.os }}-model-${{ env.REPO_ID }}-${{ env.MODEL_FILE }}
3053
- name: Install dependencies (Linux/MacOS)
3154
if: runner.os != 'Windows'
3255
run: |
3356
python -m pip install --upgrade pip
3457
python -m pip install uv
35-
RUST_LOG=trace python -m uv pip install -e .[all] --verbose
58+
python -m uv pip install -e .[all] --verbose
3659
shell: bash
37-
3860
- name: Install dependencies (Windows)
3961
if: runner.os == 'Windows'
40-
env:
41-
RUST_LOG: trace
4262
run: |
4363
python -m pip install --upgrade pip
4464
python -m pip install uv
4565
python -m uv pip install -e .[all] --verbose
46-
shell: cmd
47-
66+
shell: cmd
4867
- name: Test with pytest
4968
run: |
5069
python -m pytest
5170
5271
build-windows:
53-
72+
needs: download-model
5473
runs-on: windows-latest
5574
strategy:
5675
matrix:
5776
python-version: ["3.9", "3.10", "3.11", "3.12"]
58-
5977
steps:
6078
- uses: actions/checkout@v4
6179
with:
@@ -66,19 +84,23 @@ jobs:
6684
with:
6785
python-version: ${{ matrix.python-version }}
6886
cache: 'pip'
87+
88+
- name: Restore model cache
89+
uses: actions/cache@v3
90+
with:
91+
path: ~/.cache/huggingface/hub
92+
key: ${{ runner.os }}-model-${{ env.REPO_ID }}-${{ env.MODEL_FILE }}
6993

7094
- name: Install dependencies (Linux/MacOS)
7195
if: runner.os != 'Windows'
7296
run: |
7397
python -m pip install --upgrade pip
7498
python -m pip install uv
75-
RUST_LOG=trace python -m uv pip install -e .[all] --verbose
99+
python -m uv pip install -e .[all] --verbose
76100
shell: bash
77101

78102
- name: Install dependencies (Windows)
79103
if: runner.os == 'Windows'
80-
env:
81-
RUST_LOG: trace
82104
run: |
83105
python -m pip install --upgrade pip
84106
python -m pip install uv
@@ -90,12 +112,11 @@ jobs:
90112
python -m pytest
91113
92114
build-macos:
93-
115+
needs: download-model
94116
runs-on: macos-latest
95117
strategy:
96118
matrix:
97119
python-version: ["3.9", "3.10", "3.11", "3.12"]
98-
99120
steps:
100121
- uses: actions/checkout@v4
101122
with:
@@ -106,34 +127,36 @@ jobs:
106127
with:
107128
python-version: ${{ matrix.python-version }}
108129
cache: 'pip'
130+
131+
- name: Restore model cache
132+
uses: actions/cache@v3
133+
with:
134+
path: ~/.cache/huggingface/hub
135+
key: ${{ runner.os }}-model-${{ env.REPO_ID }}-${{ env.MODEL_FILE }}
109136

110137
- name: Install dependencies (Linux/MacOS)
111138
if: runner.os != 'Windows'
112139
run: |
113140
python -m pip install --upgrade pip
114141
python -m pip install uv
115-
RUST_LOG=trace python -m uv pip install -e .[all] --verbose
142+
python -m uv pip install -e .[all] --verbose
116143
shell: bash
117144

118145
- name: Install dependencies (Windows)
119146
if: runner.os == 'Windows'
120-
env:
121-
RUST_LOG: trace
122147
run: |
123148
python -m pip install --upgrade pip
124149
python -m pip install uv
125150
python -m uv pip install -e .[all] --verbose
126-
shell: cmd
151+
shell: cmd
127152

128153
- name: Test with pytest
129154
run: |
130155
python -m pytest
131156
132-
133157
build-macos-metal:
134-
158+
needs: download-model
135159
runs-on: macos-latest
136-
137160
steps:
138161
- uses: actions/checkout@v4
139162
with:
@@ -144,18 +167,22 @@ jobs:
144167
with:
145168
python-version: "3.9"
146169

170+
- name: Restore model cache
171+
uses: actions/cache@v3
172+
with:
173+
path: ~/.cache/huggingface/hub
174+
key: ${{ runner.os }}-model-${{ env.REPO_ID }}-${{ env.MODEL_FILE }}
175+
147176
- name: Install dependencies (Linux/MacOS)
148177
if: runner.os != 'Windows'
149178
run: |
150179
python -m pip install --upgrade pip
151180
python -m pip install uv
152-
RUST_LOG=trace CMAKE_ARGS="-DLLAMA_METAL=on" python -m uv pip install .[all] --verbose
181+
CMAKE_ARGS="-DLLAMA_METAL=on" python -m uv pip install .[all] --verbose
153182
shell: bash
154183

155184
- name: Install dependencies (Windows)
156185
if: runner.os == 'Windows'
157-
env:
158-
RUST_LOG: trace
159186
run: |
160187
python -m pip install --upgrade pip
161188
CMAKE_ARGS="-DGGML_METAL=on" python -m pip install .[all] --verbose

0 commit comments

Comments
 (0)