vertex-proxy 0.2.0__tar.gz → 0.4.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/CHANGELOG.md +21 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/PKG-INFO +49 -5
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/README.md +48 -4
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/docker-compose.yaml +5 -1
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/tests/test_config.py +44 -3
- vertex_proxy-0.4.0/tests/test_ollama.py +56 -0
- vertex_proxy-0.4.0/tests/test_openai_anthropic_bridge.py +364 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/tests/test_routing.py +90 -1
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/vertex_proxy/__init__.py +1 -1
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/vertex_proxy/config.py +22 -7
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/vertex_proxy/main.py +295 -16
- vertex_proxy-0.4.0/vertex_proxy/openai_anthropic_bridge.py +276 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/.env.example +0 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/.github/workflows/ci.yml +0 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/.github/workflows/release.yml +0 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/.gitignore +0 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/CONTRIBUTING.md +0 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/Dockerfile +0 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/LICENSE +0 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/examples/claude-code-env.sh +0 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/examples/hermes-config-snippet.yaml +0 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/examples/systemd.service +0 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/launchd/ai.hermes.vertex-proxy.plist.template +0 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/launchd/install.sh +0 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/pyproject.toml +0 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/tests/__init__.py +0 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/vertex_proxy/__main__.py +0 -0
- {vertex_proxy-0.2.0 → vertex_proxy-0.4.0}/vertex_proxy/auth.py +0 -0
|
@@ -7,6 +7,27 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
|
|
|
7
7
|
|
|
8
8
|
## [Unreleased]
|
|
9
9
|
|
|
10
|
+
## [0.4.0] - 2026-06-25
|
|
11
|
+
|
|
12
|
+
### Added
|
|
13
|
+
- Ollama backend support: route OpenAI-compatible requests to local Ollama servers via `VERTEX_PROXY_OLLAMA_BACKENDS` config. Supports wildcard (`"*"`) for auto-discovery of all models from an Ollama server's `/api/tags` endpoint. Ollama is matched after the Vertex routes so a discovered local model cannot shadow a managed model.
|
|
14
|
+
- Ollama models appear in `/v1/models` with `owned_by=ollama` and `provider=ollama`.
|
|
15
|
+
- docker-compose updated with Ollama env config and `host.docker.internal` networking.
|
|
16
|
+
|
|
17
|
+
### Fixed
|
|
18
|
+
- Strip `context_management` field from Anthropic requests before forwarding to Vertex AI. Claude Code v2.1+ sends this field, and Vertex rejects it with 400 "Extra inputs are not permitted".
|
|
19
|
+
|
|
20
|
+
## [0.3.0] - 2026-06-06
|
|
21
|
+
|
|
22
|
+
### Added
|
|
23
|
+
- OpenAI -> Anthropic bridge: Claude models are reachable through the OpenAI Chat Completions route (`/v1/chat/completions`); system/tool/assistant messages, max_tokens/temperature/stop, streaming, and tool calls are translated, and response-side Claude `tool_use` is mapped back to OpenAI `tool_calls`.
|
|
24
|
+
- Recent Anthropic models on Vertex: Opus 4.8/4.7/4.6 and Sonnet 4.6 (dateless 4.6-generation IDs).
|
|
25
|
+
- Per-provider model-listing endpoints: `/anthropic/v1/models`, `/gemini/v1/models`, `/gemini/v1beta/models`; plus OpenAI-compatible `created`/`owned_by` fields on `/v1/models`.
|
|
26
|
+
|
|
27
|
+
### Fixed
|
|
28
|
+
- Corrected the Vertex IDs for the pre-4.6 Claude family: Opus 4.5 -> `@20251101` and Haiku 4.5 -> `@20251001` (both previously 404'd at the Vertex call).
|
|
29
|
+
- The OpenAI-route Anthropic model validation now accepts the dateless 4.6-generation IDs (the old guard required an `@` and rejected `claude-opus-4-8` etc.).
|
|
30
|
+
|
|
10
31
|
## [0.2.0] - 2026-06-02
|
|
11
32
|
|
|
12
33
|
### Added
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: vertex-proxy
|
|
3
|
-
Version: 0.
|
|
3
|
+
Version: 0.4.0
|
|
4
4
|
Summary: Anthropic + Gemini + OpenAI API-compatible proxy for Google Cloud Vertex AI. Bridges static-URL API consumers to Vertex AI's service-account auth.
|
|
5
5
|
Project-URL: Homepage, https://github.com/prasadus92/vertex-proxy
|
|
6
6
|
Project-URL: Author, https://prasad.tech
|
|
@@ -116,8 +116,9 @@ curl -X POST http://127.0.0.1:8787/gemini/v1beta/models/gemini-2.5-flash:generat
|
|
|
116
116
|
|---|---|---|
|
|
117
117
|
| `POST /anthropic/v1/messages` | Anthropic Messages API | `publishers/anthropic/models/{model}:rawPredict` |
|
|
118
118
|
| `POST /gemini/v1beta/models/{m}:{action}` | Gemini generateContent API | `publishers/google/models/{m}:{action}` |
|
|
119
|
-
| `POST /openai/v1/chat/completions` | OpenAI Chat Completions | Gemini (via Vertex OpenAI-compat) + MaaS partner models (Kimi, GLM, MiniMax, Qwen, Grok) |
|
|
120
|
-
| `GET /v1/models` | - | Lists routable models |
|
|
119
|
+
| `POST /openai/v1/chat/completions` | OpenAI Chat Completions | Anthropic Claude (via OpenAI->Anthropic translation) + Gemini (via Vertex OpenAI-compat) + MaaS partner models (Kimi, GLM, MiniMax, Qwen, Grok) |
|
|
120
|
+
| `GET /v1/models` | - | Lists all routable models (OpenAI-compatible fields: id, object, created, owned_by) |
|
|
121
|
+
| `GET /anthropic/v1/models`, `/gemini/v1/models`, `/gemini/v1beta/models` | - | Per-provider model listings (for clients scoped to a single provider base_url) |
|
|
121
122
|
| `GET /health` | - | Liveness + auth check |
|
|
122
123
|
|
|
123
124
|
The OpenAI Chat Completions shape is also accepted under the `/gemini` prefix and the bare root, so clients that build their URL from a `base_url` of `.../openai`, `.../gemini`, or the server root all reach the same handler. Model-discovery probes (`/v1/models`, `/models`) are mirrored under those prefixes too.
|
|
@@ -125,14 +126,25 @@ The OpenAI Chat Completions shape is also accepted under the `/gemini` prefix an
|
|
|
125
126
|
Streaming is supported on all routes (Anthropic, Gemini, and the OpenAI-compat route).
|
|
126
127
|
Streaming requests use a no-read-timeout upstream client so long Vertex generations do not get cut off during idle periods.
|
|
127
128
|
|
|
129
|
+
Anthropic Claude is now reachable through the OpenAI Chat Completions route too: the proxy translates OpenAI <-> Anthropic Messages, so OpenAI-SDK clients can use Claude on Vertex without the Anthropic API. Tool calls are translated on the non-streaming path; streamed `tool_use` is text-only for now.
|
|
130
|
+
|
|
128
131
|
## Pre-configured models
|
|
129
132
|
|
|
130
133
|
All aliases live in [`vertex_proxy/config.py`](vertex_proxy/config.py); extend as needed.
|
|
131
134
|
|
|
132
135
|
**Anthropic** (on Vertex, `us-east5` by default)
|
|
136
|
+
|
|
137
|
+
4.6-generation IDs are dateless: the bare id IS the pinned snapshot, so the
|
|
138
|
+
proxy passes it through unchanged. Do not append `@<date>` (it 404s).
|
|
139
|
+
- `claude-opus-4-8` → `claude-opus-4-8`
|
|
140
|
+
- `claude-opus-4-7` → `claude-opus-4-7`
|
|
141
|
+
- `claude-opus-4-6` → `claude-opus-4-6`
|
|
142
|
+
- `claude-sonnet-4-6` → `claude-sonnet-4-6`
|
|
143
|
+
|
|
144
|
+
Pre-4.6 IDs carry a snapshot date that Vertex separates with `@`:
|
|
133
145
|
- `claude-sonnet-4-5-20250929` → `claude-sonnet-4-5@20250929`
|
|
134
|
-
- `claude-opus-4-5
|
|
135
|
-
- `claude-haiku-4-5
|
|
146
|
+
- `claude-opus-4-5` → `claude-opus-4-5@20251101`
|
|
147
|
+
- `claude-haiku-4-5` → `claude-haiku-4-5@20251001`
|
|
136
148
|
|
|
137
149
|
**Gemini** (on Vertex, `us-central1` by default)
|
|
138
150
|
- `gemini-2.5-pro`, `gemini-2.5-flash`, `gemini-2.0-flash`
|
|
@@ -144,6 +156,20 @@ All aliases live in [`vertex_proxy/config.py`](vertex_proxy/config.py); extend a
|
|
|
144
156
|
- `qwen3.5`, `qwen-3` (Alibaba)
|
|
145
157
|
- `grok-4.20`, `grok-4.1-fast` (xAI)
|
|
146
158
|
|
|
159
|
+
**Ollama** (local models, auto-discovered)
|
|
160
|
+
|
|
161
|
+
Set `VERTEX_PROXY_OLLAMA_BACKENDS` to a JSON object mapping model names to Ollama base URLs. Use `"*"` as a wildcard key to auto-discover all models from a server:
|
|
162
|
+
|
|
163
|
+
```bash
|
|
164
|
+
# Auto-discover all models from a local Ollama instance
|
|
165
|
+
export VERTEX_PROXY_OLLAMA_BACKENDS='{"*": "http://localhost:11434"}'
|
|
166
|
+
|
|
167
|
+
# Or map specific models to specific servers
|
|
168
|
+
export VERTEX_PROXY_OLLAMA_BACKENDS='{"qwen3:30b-a3b": "http://localhost:11434", "llama3.3:70b": "http://gpu-server:11434"}'
|
|
169
|
+
```
|
|
170
|
+
|
|
171
|
+
Discovered models appear in `/v1/models` with `owned_by=ollama` and are routed through Ollama's OpenAI-compatible endpoint (`/v1/chat/completions`).
|
|
172
|
+
|
|
147
173
|
## Recipes
|
|
148
174
|
|
|
149
175
|
### Claude Code CLI
|
|
@@ -190,6 +216,23 @@ Set the base URL environment variable the client supports (usually one of `ANTHR
|
|
|
190
216
|
export ANTHROPIC_BASE_URL=http://127.0.0.1:8787/anthropic
|
|
191
217
|
```
|
|
192
218
|
|
|
219
|
+
### OpenAI-SDK client -> Claude
|
|
220
|
+
|
|
221
|
+
Point any OpenAI-SDK client at the proxy's `/v1` base and request a Claude model by name. The proxy translates the OpenAI Chat Completions request into an Anthropic Messages call to Vertex and translates the response back:
|
|
222
|
+
|
|
223
|
+
```python
|
|
224
|
+
from openai import OpenAI
|
|
225
|
+
|
|
226
|
+
client = OpenAI(base_url="http://127.0.0.1:8787/v1", api_key="bypass")
|
|
227
|
+
resp = client.chat.completions.create(
|
|
228
|
+
model="claude-opus-4-8",
|
|
229
|
+
messages=[{"role": "user", "content": "hello"}],
|
|
230
|
+
)
|
|
231
|
+
print(resp.choices[0].message.content)
|
|
232
|
+
```
|
|
233
|
+
|
|
234
|
+
System messages, `max_tokens`/`temperature`/`stop`, streaming, and tool calls are translated (streamed `tool_use` is not yet).
|
|
235
|
+
|
|
193
236
|
## Run as a service (macOS launchd)
|
|
194
237
|
|
|
195
238
|
```bash
|
|
@@ -219,6 +262,7 @@ All settings accept `VERTEX_PROXY_` env var prefix or CLI flags.
|
|
|
219
262
|
| `VERTEX_PROXY_MAAS_REGION` | `us-central1` | Region for Kimi / GLM / MiniMax / Qwen / Grok |
|
|
220
263
|
| `VERTEX_PROXY_HOST` | `127.0.0.1` | Bind host |
|
|
221
264
|
| `VERTEX_PROXY_PORT` | `8787` | Bind port |
|
|
265
|
+
| `VERTEX_PROXY_OLLAMA_BACKENDS` | `{}` | JSON map of model names to Ollama base URLs; `"*"` auto-discovers |
|
|
222
266
|
| `VERTEX_PROXY_TOKEN_REFRESH_SECONDS` | `3000` | Token refresh interval (50 min) |
|
|
223
267
|
| `VERTEX_PROXY_LOG_LEVEL` | `info` | uvicorn log level |
|
|
224
268
|
|
|
@@ -84,8 +84,9 @@ curl -X POST http://127.0.0.1:8787/gemini/v1beta/models/gemini-2.5-flash:generat
|
|
|
84
84
|
|---|---|---|
|
|
85
85
|
| `POST /anthropic/v1/messages` | Anthropic Messages API | `publishers/anthropic/models/{model}:rawPredict` |
|
|
86
86
|
| `POST /gemini/v1beta/models/{m}:{action}` | Gemini generateContent API | `publishers/google/models/{m}:{action}` |
|
|
87
|
-
| `POST /openai/v1/chat/completions` | OpenAI Chat Completions | Gemini (via Vertex OpenAI-compat) + MaaS partner models (Kimi, GLM, MiniMax, Qwen, Grok) |
|
|
88
|
-
| `GET /v1/models` | - | Lists routable models |
|
|
87
|
+
| `POST /openai/v1/chat/completions` | OpenAI Chat Completions | Anthropic Claude (via OpenAI->Anthropic translation) + Gemini (via Vertex OpenAI-compat) + MaaS partner models (Kimi, GLM, MiniMax, Qwen, Grok) |
|
|
88
|
+
| `GET /v1/models` | - | Lists all routable models (OpenAI-compatible fields: id, object, created, owned_by) |
|
|
89
|
+
| `GET /anthropic/v1/models`, `/gemini/v1/models`, `/gemini/v1beta/models` | - | Per-provider model listings (for clients scoped to a single provider base_url) |
|
|
89
90
|
| `GET /health` | - | Liveness + auth check |
|
|
90
91
|
|
|
91
92
|
The OpenAI Chat Completions shape is also accepted under the `/gemini` prefix and the bare root, so clients that build their URL from a `base_url` of `.../openai`, `.../gemini`, or the server root all reach the same handler. Model-discovery probes (`/v1/models`, `/models`) are mirrored under those prefixes too.
|
|
@@ -93,14 +94,25 @@ The OpenAI Chat Completions shape is also accepted under the `/gemini` prefix an
|
|
|
93
94
|
Streaming is supported on all routes (Anthropic, Gemini, and the OpenAI-compat route).
|
|
94
95
|
Streaming requests use a no-read-timeout upstream client so long Vertex generations do not get cut off during idle periods.
|
|
95
96
|
|
|
97
|
+
Anthropic Claude is now reachable through the OpenAI Chat Completions route too: the proxy translates OpenAI <-> Anthropic Messages, so OpenAI-SDK clients can use Claude on Vertex without the Anthropic API. Tool calls are translated on the non-streaming path; streamed `tool_use` is text-only for now.
|
|
98
|
+
|
|
96
99
|
## Pre-configured models
|
|
97
100
|
|
|
98
101
|
All aliases live in [`vertex_proxy/config.py`](vertex_proxy/config.py); extend as needed.
|
|
99
102
|
|
|
100
103
|
**Anthropic** (on Vertex, `us-east5` by default)
|
|
104
|
+
|
|
105
|
+
4.6-generation IDs are dateless: the bare id IS the pinned snapshot, so the
|
|
106
|
+
proxy passes it through unchanged. Do not append `@<date>` (it 404s).
|
|
107
|
+
- `claude-opus-4-8` → `claude-opus-4-8`
|
|
108
|
+
- `claude-opus-4-7` → `claude-opus-4-7`
|
|
109
|
+
- `claude-opus-4-6` → `claude-opus-4-6`
|
|
110
|
+
- `claude-sonnet-4-6` → `claude-sonnet-4-6`
|
|
111
|
+
|
|
112
|
+
Pre-4.6 IDs carry a snapshot date that Vertex separates with `@`:
|
|
101
113
|
- `claude-sonnet-4-5-20250929` → `claude-sonnet-4-5@20250929`
|
|
102
|
-
- `claude-opus-4-5
|
|
103
|
-
- `claude-haiku-4-5
|
|
114
|
+
- `claude-opus-4-5` → `claude-opus-4-5@20251101`
|
|
115
|
+
- `claude-haiku-4-5` → `claude-haiku-4-5@20251001`
|
|
104
116
|
|
|
105
117
|
**Gemini** (on Vertex, `us-central1` by default)
|
|
106
118
|
- `gemini-2.5-pro`, `gemini-2.5-flash`, `gemini-2.0-flash`
|
|
@@ -112,6 +124,20 @@ All aliases live in [`vertex_proxy/config.py`](vertex_proxy/config.py); extend a
|
|
|
112
124
|
- `qwen3.5`, `qwen-3` (Alibaba)
|
|
113
125
|
- `grok-4.20`, `grok-4.1-fast` (xAI)
|
|
114
126
|
|
|
127
|
+
**Ollama** (local models, auto-discovered)
|
|
128
|
+
|
|
129
|
+
Set `VERTEX_PROXY_OLLAMA_BACKENDS` to a JSON object mapping model names to Ollama base URLs. Use `"*"` as a wildcard key to auto-discover all models from a server:
|
|
130
|
+
|
|
131
|
+
```bash
|
|
132
|
+
# Auto-discover all models from a local Ollama instance
|
|
133
|
+
export VERTEX_PROXY_OLLAMA_BACKENDS='{"*": "http://localhost:11434"}'
|
|
134
|
+
|
|
135
|
+
# Or map specific models to specific servers
|
|
136
|
+
export VERTEX_PROXY_OLLAMA_BACKENDS='{"qwen3:30b-a3b": "http://localhost:11434", "llama3.3:70b": "http://gpu-server:11434"}'
|
|
137
|
+
```
|
|
138
|
+
|
|
139
|
+
Discovered models appear in `/v1/models` with `owned_by=ollama` and are routed through Ollama's OpenAI-compatible endpoint (`/v1/chat/completions`).
|
|
140
|
+
|
|
115
141
|
## Recipes
|
|
116
142
|
|
|
117
143
|
### Claude Code CLI
|
|
@@ -158,6 +184,23 @@ Set the base URL environment variable the client supports (usually one of `ANTHR
|
|
|
158
184
|
export ANTHROPIC_BASE_URL=http://127.0.0.1:8787/anthropic
|
|
159
185
|
```
|
|
160
186
|
|
|
187
|
+
### OpenAI-SDK client -> Claude
|
|
188
|
+
|
|
189
|
+
Point any OpenAI-SDK client at the proxy's `/v1` base and request a Claude model by name. The proxy translates the OpenAI Chat Completions request into an Anthropic Messages call to Vertex and translates the response back:
|
|
190
|
+
|
|
191
|
+
```python
|
|
192
|
+
from openai import OpenAI
|
|
193
|
+
|
|
194
|
+
client = OpenAI(base_url="http://127.0.0.1:8787/v1", api_key="bypass")
|
|
195
|
+
resp = client.chat.completions.create(
|
|
196
|
+
model="claude-opus-4-8",
|
|
197
|
+
messages=[{"role": "user", "content": "hello"}],
|
|
198
|
+
)
|
|
199
|
+
print(resp.choices[0].message.content)
|
|
200
|
+
```
|
|
201
|
+
|
|
202
|
+
System messages, `max_tokens`/`temperature`/`stop`, streaming, and tool calls are translated (streamed `tool_use` is not yet).
|
|
203
|
+
|
|
161
204
|
## Run as a service (macOS launchd)
|
|
162
205
|
|
|
163
206
|
```bash
|
|
@@ -187,6 +230,7 @@ All settings accept `VERTEX_PROXY_` env var prefix or CLI flags.
|
|
|
187
230
|
| `VERTEX_PROXY_MAAS_REGION` | `us-central1` | Region for Kimi / GLM / MiniMax / Qwen / Grok |
|
|
188
231
|
| `VERTEX_PROXY_HOST` | `127.0.0.1` | Bind host |
|
|
189
232
|
| `VERTEX_PROXY_PORT` | `8787` | Bind port |
|
|
233
|
+
| `VERTEX_PROXY_OLLAMA_BACKENDS` | `{}` | JSON map of model names to Ollama base URLs; `"*"` auto-discovers |
|
|
190
234
|
| `VERTEX_PROXY_TOKEN_REFRESH_SECONDS` | `3000` | Token refresh interval (50 min) |
|
|
191
235
|
| `VERTEX_PROXY_LOG_LEVEL` | `info` | uvicorn log level |
|
|
192
236
|
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
#
|
|
1
|
+
# vertex-proxy docker-compose setup.
|
|
2
2
|
# Usage:
|
|
3
3
|
# 1. Copy .env.example to .env and fill in your values
|
|
4
4
|
# 2. Place your GCP service-account JSON at ./credentials/gcp-key.json
|
|
@@ -22,8 +22,12 @@ services:
|
|
|
22
22
|
VERTEX_PROXY_ANTHROPIC_REGION: ${VERTEX_PROXY_ANTHROPIC_REGION:-us-east5}
|
|
23
23
|
VERTEX_PROXY_GEMINI_REGION: ${VERTEX_PROXY_GEMINI_REGION:-us-central1}
|
|
24
24
|
VERTEX_PROXY_LOG_LEVEL: ${VERTEX_PROXY_LOG_LEVEL:-info}
|
|
25
|
+
# Ollama backend: auto-discover all models from the host's Ollama
|
|
26
|
+
VERTEX_PROXY_OLLAMA_BACKENDS: '{"*": "http://host.docker.internal:11434"}'
|
|
25
27
|
# Uncomment to require a bearer token on all routes:
|
|
26
28
|
# VERTEX_PROXY_API_KEY: ${VERTEX_PROXY_API_KEY}
|
|
29
|
+
extra_hosts:
|
|
30
|
+
- "host.docker.internal:host-gateway"
|
|
27
31
|
healthcheck:
|
|
28
32
|
test: ["CMD", "python", "-c", "import urllib.request,sys;sys.exit(0 if urllib.request.urlopen('http://127.0.0.1:8787/health',timeout=3).status==200 else 1)"]
|
|
29
33
|
interval: 30s
|
|
@@ -2,6 +2,8 @@
|
|
|
2
2
|
|
|
3
3
|
from __future__ import annotations
|
|
4
4
|
|
|
5
|
+
import re
|
|
6
|
+
|
|
5
7
|
from vertex_proxy.config import Settings
|
|
6
8
|
|
|
7
9
|
|
|
@@ -20,11 +22,50 @@ def test_anthropic_aliases_include_sonnet() -> None:
|
|
|
20
22
|
assert s.anthropic_model_aliases["claude-sonnet-4-5-20250929"] == "claude-sonnet-4-5@20250929"
|
|
21
23
|
|
|
22
24
|
|
|
23
|
-
|
|
24
|
-
|
|
25
|
+
# Vertex uses two ID conventions, split at the Claude 4.6 generation:
|
|
26
|
+
# * 4.6-gen and later (Opus 4.6/4.7/4.8, Sonnet 4.6) use a DATELESS bare ID;
|
|
27
|
+
# appending '@<date>' makes the Vertex call 404.
|
|
28
|
+
# * pre-4.6 (Sonnet 4.5/4, Opus 4.5, Haiku 4.5) carry a snapshot date that
|
|
29
|
+
# Vertex separates with '@', e.g. claude-haiku-4-5@20251001.
|
|
30
|
+
_DATELESS_46_GEN = {
|
|
31
|
+
"claude-opus-4-8",
|
|
32
|
+
"claude-opus-4-7",
|
|
33
|
+
"claude-opus-4-6",
|
|
34
|
+
"claude-sonnet-4-6",
|
|
35
|
+
}
|
|
36
|
+
|
|
37
|
+
|
|
38
|
+
def test_anthropic_46_gen_aliases_are_dateless() -> None:
|
|
39
|
+
"""4.6-generation Vertex IDs are bare (no '@', no date); a suffix 404s."""
|
|
40
|
+
s = Settings()
|
|
41
|
+
for model in _DATELESS_46_GEN:
|
|
42
|
+
vertex_id = s.anthropic_model_aliases[model]
|
|
43
|
+
assert vertex_id == model, f"4.6-gen alias {model!r} must map to bare id, got {vertex_id!r}"
|
|
44
|
+
assert "@" not in vertex_id, (
|
|
45
|
+
f"4.6-gen alias {model!r} must NOT carry '@', got {vertex_id!r}"
|
|
46
|
+
)
|
|
47
|
+
|
|
48
|
+
|
|
49
|
+
def test_anthropic_pre_46_aliases_use_at_sign_date() -> None:
|
|
50
|
+
"""Pre-4.6 Vertex IDs separate the snapshot date with '@', never '-'."""
|
|
25
51
|
s = Settings()
|
|
26
52
|
for alias, vertex_id in s.anthropic_model_aliases.items():
|
|
27
|
-
|
|
53
|
+
if vertex_id in _DATELESS_46_GEN:
|
|
54
|
+
continue
|
|
55
|
+
assert "@" in vertex_id, f"pre-4.6 alias {alias!r} → {vertex_id!r} missing '@' date sep"
|
|
56
|
+
# The eight digits after '@' must be the snapshot date.
|
|
57
|
+
head, _, date = vertex_id.partition("@")
|
|
58
|
+
assert date.isdigit() and len(date) == 8, f"{alias!r} → {vertex_id!r} bad date segment"
|
|
59
|
+
# The head (model name) must NOT carry a '-YYYYMMDD' dash-date; on Vertex
|
|
60
|
+
# the date is always '@'-separated, never '-' (that is the Claude-API form).
|
|
61
|
+
assert not re.search(r"-\d{8}$", head), f"{alias!r} has a '-' date in {vertex_id!r}"
|
|
62
|
+
|
|
63
|
+
|
|
64
|
+
def test_opus_45_and_haiku_45_use_correct_snapshot_dates() -> None:
|
|
65
|
+
"""Opus 4.5 = @20251101 and Haiku 4.5 = @20251001 on Vertex (not @20250929)."""
|
|
66
|
+
s = Settings()
|
|
67
|
+
assert s.anthropic_model_aliases["claude-opus-4-5"] == "claude-opus-4-5@20251101"
|
|
68
|
+
assert s.anthropic_model_aliases["claude-haiku-4-5"] == "claude-haiku-4-5@20251001"
|
|
28
69
|
|
|
29
70
|
|
|
30
71
|
def test_gemini_aliases_include_pro_and_flash() -> None:
|
|
@@ -0,0 +1,56 @@
|
|
|
1
|
+
"""Unit tests for Ollama backend integration."""
|
|
2
|
+
|
|
3
|
+
from __future__ import annotations
|
|
4
|
+
|
|
5
|
+
import pytest
|
|
6
|
+
from httpx import ASGITransport, AsyncClient
|
|
7
|
+
|
|
8
|
+
from vertex_proxy.config import Settings
|
|
9
|
+
from vertex_proxy.main import build_app
|
|
10
|
+
|
|
11
|
+
|
|
12
|
+
def test_ollama_backends_default_empty() -> None:
|
|
13
|
+
"""Ollama backends default to an empty dict."""
|
|
14
|
+
s = Settings()
|
|
15
|
+
assert s.ollama_backends == {}
|
|
16
|
+
|
|
17
|
+
|
|
18
|
+
def test_ollama_backends_explicit_model() -> None:
|
|
19
|
+
"""Explicit model-to-URL mapping is parsed correctly."""
|
|
20
|
+
s = Settings(ollama_backends={"qwen3:30b": "http://localhost:11434"})
|
|
21
|
+
assert s.ollama_backends["qwen3:30b"] == "http://localhost:11434"
|
|
22
|
+
|
|
23
|
+
|
|
24
|
+
def test_ollama_backends_wildcard() -> None:
|
|
25
|
+
"""Wildcard key '*' is accepted in config."""
|
|
26
|
+
s = Settings(ollama_backends={"*": "http://gpu:11434"})
|
|
27
|
+
assert s.ollama_backends["*"] == "http://gpu:11434"
|
|
28
|
+
|
|
29
|
+
|
|
30
|
+
@pytest.mark.anyio
|
|
31
|
+
async def test_ollama_models_in_v1_models() -> None:
|
|
32
|
+
"""Ollama models registered via explicit config appear in /v1/models."""
|
|
33
|
+
settings = Settings(
|
|
34
|
+
ollama_backends={"test-model:7b": "http://fake:11434"},
|
|
35
|
+
)
|
|
36
|
+
app = build_app(settings)
|
|
37
|
+
|
|
38
|
+
# Manually populate the Ollama registry (skip async discovery)
|
|
39
|
+
from vertex_proxy import main as main_mod
|
|
40
|
+
|
|
41
|
+
main_mod._OLLAMA_MODELS["test-model:7b"] = "http://fake:11434"
|
|
42
|
+
|
|
43
|
+
try:
|
|
44
|
+
transport = ASGITransport(app=app)
|
|
45
|
+
async with AsyncClient(transport=transport, base_url="http://test") as client:
|
|
46
|
+
resp = await client.get("/v1/models")
|
|
47
|
+
assert resp.status_code == 200
|
|
48
|
+
data = resp.json()
|
|
49
|
+
model_ids = [m["id"] for m in data["data"]]
|
|
50
|
+
assert "test-model:7b" in model_ids
|
|
51
|
+
# Check owned_by and provider
|
|
52
|
+
ollama_model = [m for m in data["data"] if m["id"] == "test-model:7b"][0]
|
|
53
|
+
assert ollama_model["owned_by"] == "ollama"
|
|
54
|
+
assert ollama_model["provider"] == "ollama"
|
|
55
|
+
finally:
|
|
56
|
+
main_mod._OLLAMA_MODELS.clear()
|