hud-python 0.2.4__py3-none-any.whl → 0.2.6__py3-none-any.whl
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.
Potentially problematic release.
This version of hud-python might be problematic. Click here for more details.
- hud/__init__.py +22 -2
- hud/adapters/claude/adapter.py +9 -2
- hud/adapters/claude/tests/__init__.py +1 -0
- hud/adapters/claude/tests/test_adapter.py +519 -0
- hud/adapters/common/types.py +5 -1
- hud/adapters/operator/adapter.py +4 -0
- hud/adapters/operator/tests/__init__.py +1 -0
- hud/adapters/operator/tests/test_adapter.py +370 -0
- hud/agent/__init__.py +4 -0
- hud/agent/base.py +18 -2
- hud/agent/claude.py +20 -17
- hud/agent/claude_plays_pokemon.py +283 -0
- hud/agent/langchain.py +12 -7
- hud/agent/misc/__init__.py +3 -0
- hud/agent/misc/response_agent.py +80 -0
- hud/agent/operator.py +27 -19
- hud/agent/tests/__init__.py +1 -0
- hud/agent/tests/test_base.py +202 -0
- hud/env/docker_client.py +28 -18
- hud/env/environment.py +32 -16
- hud/env/local_docker_client.py +83 -42
- hud/env/remote_client.py +1 -3
- hud/env/remote_docker_client.py +71 -14
- hud/exceptions.py +12 -0
- hud/gym.py +71 -53
- hud/job.py +59 -14
- hud/server/requests.py +26 -4
- hud/settings.py +7 -1
- hud/task.py +45 -33
- hud/taskset.py +56 -4
- hud/telemetry/__init__.py +21 -0
- hud/telemetry/_trace.py +173 -0
- hud/telemetry/context.py +169 -0
- hud/telemetry/exporter.py +417 -0
- hud/telemetry/instrumentation/__init__.py +3 -0
- hud/telemetry/instrumentation/mcp.py +495 -0
- hud/telemetry/instrumentation/registry.py +59 -0
- hud/telemetry/mcp_models.py +331 -0
- hud/telemetry/tests/__init__.py +1 -0
- hud/telemetry/tests/test_context.py +207 -0
- hud/telemetry/tests/test_trace.py +270 -0
- hud/types.py +11 -27
- hud/utils/common.py +22 -2
- hud/utils/misc.py +53 -0
- hud/utils/tests/test_version.py +1 -1
- hud/version.py +7 -0
- {hud_python-0.2.4.dist-info → hud_python-0.2.6.dist-info}/METADATA +98 -30
- hud_python-0.2.6.dist-info/RECORD +84 -0
- hud_python-0.2.4.dist-info/RECORD +0 -62
- {hud_python-0.2.4.dist-info → hud_python-0.2.6.dist-info}/WHEEL +0 -0
- {hud_python-0.2.4.dist-info → hud_python-0.2.6.dist-info}/licenses/LICENSE +0 -0
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: hud-python
|
|
3
|
-
Version: 0.2.
|
|
3
|
+
Version: 0.2.6
|
|
4
4
|
Summary: SDK for the HUD evaluation platform.
|
|
5
5
|
Project-URL: Homepage, https://github.com/hud-evals/hud-sdk
|
|
6
6
|
Project-URL: Bug Tracker, https://github.com/hud-evals/hud-sdk/issues
|
|
@@ -38,11 +38,13 @@ Classifier: Programming Language :: Python :: 3.13
|
|
|
38
38
|
Requires-Python: <3.14,>=3.10
|
|
39
39
|
Requires-Dist: aiodocker>=0.24.0
|
|
40
40
|
Requires-Dist: anthropic
|
|
41
|
+
Requires-Dist: dotenv>=0.9.9
|
|
41
42
|
Requires-Dist: httpx<1,>=0.23.0
|
|
42
43
|
Requires-Dist: inspect-ai>=0.3.80
|
|
43
44
|
Requires-Dist: ipykernel
|
|
44
45
|
Requires-Dist: langchain
|
|
45
46
|
Requires-Dist: langchain-openai
|
|
47
|
+
Requires-Dist: mcp
|
|
46
48
|
Requires-Dist: numpy
|
|
47
49
|
Requires-Dist: openai
|
|
48
50
|
Requires-Dist: pillow>=11.1.0
|
|
@@ -50,6 +52,7 @@ Requires-Dist: pydantic-settings<3,>=2
|
|
|
50
52
|
Requires-Dist: pydantic<3,>=2
|
|
51
53
|
Requires-Dist: textdistance<5,>=4.5.0
|
|
52
54
|
Requires-Dist: toml>=0.10.2
|
|
55
|
+
Requires-Dist: wrapt>=1.14.0
|
|
53
56
|
Provides-Extra: dev
|
|
54
57
|
Requires-Dist: anthropic; extra == 'dev'
|
|
55
58
|
Requires-Dist: dotenv; extra == 'dev'
|
|
@@ -66,25 +69,67 @@ Requires-Dist: pytest<9,>=8.1.1; extra == 'dev'
|
|
|
66
69
|
Requires-Dist: ruff==0.11.8; extra == 'dev'
|
|
67
70
|
Description-Content-Type: text/markdown
|
|
68
71
|
|
|
69
|
-
|
|
72
|
+
<div align="left">
|
|
73
|
+
<img src="https://raw.githubusercontent.com/hud-evals/hud-sdk/main/docs/logo/hud_logo.svg" alt="HUD" width="150" style="margin-bottom: 20px;"/>
|
|
74
|
+
</div>
|
|
70
75
|
|
|
71
|
-
|
|
76
|
+
<h3>
|
|
77
|
+
Evaluate your Computer Use AI agents across web browsers, desktop environments, and custom scenarios.
|
|
78
|
+
</h3>
|
|
72
79
|
|
|
73
|
-
|
|
80
|
+
### 🚀 Are you a startup building agents?
|
|
81
|
+
|
|
82
|
+
[📅 Hop on a call](https://cal.com/jay-ram-z6st6w/demo) or [📧 founders@hud.so](mailto:founders@hud.so)
|
|
83
|
+
|
|
84
|
+
We're here to help with eval strategies, custom environments, or improving your agent architecture!
|
|
85
|
+
|
|
86
|
+
|
|
87
|
+
> **Early Release Notice**: We'd love to hear your feedback in [Issues](https://github.com/hud-evals/hud-sdk/issues), as the SDK is still evolving!
|
|
74
88
|
|
|
75
89
|
[](https://pypi.org/project/hud-python/)
|
|
76
90
|
|
|
77
|
-
|
|
91
|
+
## ✨ What You Can Do
|
|
78
92
|
|
|
79
|
-
|
|
93
|
+
**Evaluate Existing Benchmarks**
|
|
94
|
+
```python
|
|
95
|
+
from hud import load_taskset, run_job, ClaudeAgent
|
|
80
96
|
|
|
81
|
-
|
|
97
|
+
taskset = await load_taskset("WebVoyager") # or GAIA, OSWorld-Ubuntu, Mind2Web
|
|
98
|
+
job = await run_job(ClaudeAgent, taskset, "my-evaluation")
|
|
99
|
+
```
|
|
82
100
|
|
|
83
|
-
|
|
84
|
-
|
|
101
|
+
**Create Custom Tasks**
|
|
102
|
+
```python
|
|
103
|
+
from hud.task import Task
|
|
85
104
|
|
|
86
|
-
|
|
87
|
-
|
|
105
|
+
task = Task(
|
|
106
|
+
prompt="Find and book the cheapest flight from NYC to Paris",
|
|
107
|
+
gym="hud-browser",
|
|
108
|
+
setup=("goto", "https://kayak.com"),
|
|
109
|
+
evaluate=("page_contains", "confirmation")
|
|
110
|
+
)
|
|
111
|
+
```
|
|
112
|
+
|
|
113
|
+
**Build Custom Environments**
|
|
114
|
+
```python
|
|
115
|
+
from hud.types import CustomGym
|
|
116
|
+
|
|
117
|
+
# Launch any website as an environment
|
|
118
|
+
custom_gym = CustomGym(
|
|
119
|
+
image_or_build_context="nginx:alpine",
|
|
120
|
+
location="local"
|
|
121
|
+
)
|
|
122
|
+
|
|
123
|
+
# Or create complex Docker environments - see environments/ folder for examples
|
|
124
|
+
```
|
|
125
|
+
|
|
126
|
+
**Trace Tool Calls Alongside HUD Environments (or Independently)**
|
|
127
|
+
```python
|
|
128
|
+
import hud
|
|
129
|
+
|
|
130
|
+
with hud.trace("my-agent-run"):
|
|
131
|
+
# Your agent code here - MCP calls automatically captured
|
|
132
|
+
result = await agent.run(task)
|
|
88
133
|
```
|
|
89
134
|
|
|
90
135
|
## Quick Start
|
|
@@ -95,6 +140,17 @@ export HUD_API_KEY=your_api_key_here
|
|
|
95
140
|
pip install hud-python
|
|
96
141
|
```
|
|
97
142
|
|
|
143
|
+
### API Key Setup
|
|
144
|
+
|
|
145
|
+
Before getting started, you'll need to obtain an API key:
|
|
146
|
+
|
|
147
|
+
1. Visit [app.hud.so](https://app.hud.so) to create a free account and generate your API key
|
|
148
|
+
2. Set it in your environment or .env file:
|
|
149
|
+
|
|
150
|
+
```bash
|
|
151
|
+
export HUD_API_KEY=your_api_key_here
|
|
152
|
+
```
|
|
153
|
+
|
|
98
154
|
### Simple Browser Example with Claude Computer Use
|
|
99
155
|
|
|
100
156
|
> This example uses the `@register_job("test-run")` decorator, so the results of this run will appear under the job named "test-run" on the your [HUD Jobs page](https://app.hud.so/jobs).
|
|
@@ -119,7 +175,7 @@ async def main():
|
|
|
119
175
|
# Create environment using the gym module
|
|
120
176
|
env = await gym.make(task)
|
|
121
177
|
|
|
122
|
-
# Initialize
|
|
178
|
+
# Initialize Claude agent (API key is loaded automatically)
|
|
123
179
|
agent = ClaudeAgent()
|
|
124
180
|
|
|
125
181
|
# Agent loop with predict and step functions
|
|
@@ -137,7 +193,6 @@ async def main():
|
|
|
137
193
|
|
|
138
194
|
if __name__ == "__main__":
|
|
139
195
|
asyncio.run(main())
|
|
140
|
-
|
|
141
196
|
```
|
|
142
197
|
|
|
143
198
|
Alternatively, run a full evaluation set via the ```run_job``` command:
|
|
@@ -145,32 +200,45 @@ Alternatively, run a full evaluation set via the ```run_job``` command:
|
|
|
145
200
|
```python
|
|
146
201
|
from hud import load_taskset, run_job, ClaudeAgent
|
|
147
202
|
|
|
148
|
-
#
|
|
203
|
+
# Load a benchmark
|
|
149
204
|
taskset = load_taskset("GAIA")
|
|
150
205
|
|
|
151
|
-
#
|
|
206
|
+
# Evaluate
|
|
152
207
|
job = await run_job(ClaudeAgent, taskset, "test-gaia-job")
|
|
153
208
|
|
|
154
|
-
#
|
|
209
|
+
# Get results OR view them in app.hud.so
|
|
155
210
|
print(await job.get_analytics())
|
|
156
211
|
```
|
|
157
212
|
|
|
213
|
+
## Ready-to-Use TaskSets
|
|
214
|
+
|
|
215
|
+
- **WebVoyager** - Web navigation and interaction
|
|
216
|
+
- **Mind2Web** - Complex web application tasks
|
|
217
|
+
- **GAIA** - Question answering and reasoning
|
|
218
|
+
- **OSWorld-Ubuntu** - Desktop interaction
|
|
219
|
+
- **hud-samples** - Getting started examples
|
|
220
|
+
|
|
221
|
+
## Community
|
|
222
|
+
|
|
223
|
+
**Contributing Custom Environments**
|
|
224
|
+
|
|
225
|
+
Add your environment to the `environments/` folder and submit a PR! Examples:
|
|
226
|
+
- `environments/novnc_ubuntu/` - Ubuntu with VNC access
|
|
227
|
+
- `environments/pokemon_controller/` - Pokemon emulator environment (In Development)
|
|
228
|
+
- `environments/qa_controller/` - Lightweight app sample
|
|
229
|
+
|
|
230
|
+
See [Custom Environments Guide](https://docs.hud.so/environment-creation) for details.
|
|
231
|
+
|
|
158
232
|
## Documentation Sections
|
|
159
233
|
|
|
160
234
|
Explore the core concepts and features of the SDK:
|
|
161
235
|
|
|
162
|
-
* **[
|
|
163
|
-
* **[Environments](https://
|
|
164
|
-
* **[Agents](https://
|
|
165
|
-
* **[
|
|
166
|
-
* **[
|
|
167
|
-
* **[
|
|
168
|
-
* **Advanced Topics**:
|
|
169
|
-
* **[CLA Action Details](https://documentation.hud.so/advanced/cla-details)**: Explore the standardized action format.
|
|
170
|
-
* **[Custom Environments](https://documentation.hud.so/advanced/custom-environments)**: Build your own Docker-based local or remote environments.
|
|
171
|
-
* **[Advanced Environment Control](https://documentation.hud.so/advanced/environment-control)**: Use `invoke`, `execute`, and `_setup` for finer control.
|
|
172
|
-
|
|
173
|
-
* **[Full API Reference](https://documentation.hud.so/api-reference/gym)**: Detailed specifications for all modules and classes.
|
|
236
|
+
* **[Task Creation](https://docs.hud.so/task-creation)**: Build custom evaluation scenarios with setup and evaluation criteria.
|
|
237
|
+
* **[Environments](https://docs.hud.so/environments/browser)**: Understand browser environments and create custom Docker-based environments.
|
|
238
|
+
* **[Agents](https://docs.hud.so/concepts/agent)**: Learn about the agent architecture (Claude, Operator) and how they process observations and predict actions.
|
|
239
|
+
* **[Jobs](https://docs.hud.so/concepts/job)**: Group related runs for analysis and viewing on the HUD platform.
|
|
240
|
+
* **[MCP Telemetry](https://docs.hud.so/telemetry/mcp)**: Automatic tracing of Model Context Protocol interactions.
|
|
241
|
+
* **[Full API Reference](https://docs.hud.so/api-reference/gym)**: Detailed specifications for all modules and classes.
|
|
174
242
|
|
|
175
243
|
## [Examples](examples/)
|
|
176
244
|
|
|
@@ -183,7 +251,7 @@ We recommend you first take a look at the example notebooks showing how to use t
|
|
|
183
251
|
|
|
184
252
|
## Documentation
|
|
185
253
|
|
|
186
|
-
For comprehensive guides, examples, and API reference, visit [our docs](https://
|
|
254
|
+
For comprehensive guides, examples, and API reference, visit [our docs](https://docs.hud.so/introduction)
|
|
187
255
|
|
|
188
256
|
## License
|
|
189
257
|
|
|
@@ -201,4 +269,4 @@ If you use this SDK in your research, please cite it as follows:
|
|
|
201
269
|
url = {https://github.com/hud-evals/hud-sdk},
|
|
202
270
|
langid = {en}
|
|
203
271
|
}
|
|
204
|
-
```
|
|
272
|
+
```
|
|
@@ -0,0 +1,84 @@
|
|
|
1
|
+
hud/__init__.py,sha256=6PlxwtjYyaqk6UAyHLJZhsiHRlgndH-Jja9f9BtInUY,1063
|
|
2
|
+
hud/exceptions.py,sha256=pifKvSqxj9_g4NfARVyH5a-lTThhi9XW06tIXaBakQw,5526
|
|
3
|
+
hud/gym.py,sha256=Dl7nur2QTxoVNAcWIvFjuGAbKmoc7CVgjV5gWd35usU,4544
|
|
4
|
+
hud/job.py,sha256=bd88L83L3uqdXE7B3Bjsk8hGk95OggJiLjItFsZXDoQ,25116
|
|
5
|
+
hud/settings.py,sha256=3zALwVbPTaDc01-dR_-rGsrDfc-ieMIcmO5avv6S2Y0,1510
|
|
6
|
+
hud/task.py,sha256=AMmJLYl3BjX8TfBY4ZuR_QIXhTkWDX-4C_Pbi3HziVg,5505
|
|
7
|
+
hud/taskset.py,sha256=ou2Ivulv392txtDvXDpvJAgHj_4h2LZNBaxpeC_mRnw,4903
|
|
8
|
+
hud/trajectory.py,sha256=OrcRbxK_ejFp1VhJCjZnM1WCmCXxEOK4CxNjCngcsjo,3721
|
|
9
|
+
hud/types.py,sha256=O8eotDLw4onwRreX9xLPZ2T11qf8wwUidaqGI7jvagY,1808
|
|
10
|
+
hud/version.py,sha256=4c7HS3iYXZGe4vuaPb8pgYwVInf3F3G95h0PITpm6aw,104
|
|
11
|
+
hud/adapters/__init__.py,sha256=zz24KdC_e9TJPgWo6y57_8SzevEE5ak4Cm6tXzMxwRk,266
|
|
12
|
+
hud/adapters/claude/__init__.py,sha256=i7QEF-29FLb9qxp1eYtXs-adIk_tG54tL-9g6d3xodk,100
|
|
13
|
+
hud/adapters/claude/adapter.py,sha256=_qUD0iu0_Y_8yuhrsZw2E5wNv8RB-Aa7BqclAmNHdtI,6096
|
|
14
|
+
hud/adapters/claude/tests/__init__.py,sha256=9GZj0rz4tTkiPnLfxTmyBPr-s8UZc3gph6WH8fs8T34,39
|
|
15
|
+
hud/adapters/claude/tests/test_adapter.py,sha256=cAdHEoqLngLiV7QwlWJ0KuNgb1vNv9WZTPQMnxhMDKI,18319
|
|
16
|
+
hud/adapters/common/__init__.py,sha256=BjdZWJVs_AKtpFrt-tNsdQRjnz7D97DFEQirJ-r0mp8,118
|
|
17
|
+
hud/adapters/common/adapter.py,sha256=fxdz1S6uwtHczLoDzmbuT0icQ49ywGMTiuVv8oJL4g4,5822
|
|
18
|
+
hud/adapters/common/types.py,sha256=UpY-Up5s53ujzKdEp6SLf0E2tFS7iQYE0_IZRZrILxo,5067
|
|
19
|
+
hud/adapters/common/tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
20
|
+
hud/adapters/common/tests/test_adapter.py,sha256=rTD36LjvytHqMIyOLDyrn0RLIkd20s6f6dwoBEarJaw,8744
|
|
21
|
+
hud/adapters/operator/__init__.py,sha256=31vTRs268_TOLd-TeQRKau5bDYy78wxCNpJFhD5_l8U,104
|
|
22
|
+
hud/adapters/operator/adapter.py,sha256=pl-uBLcPVyFFASnq6RtW8PbPscTre7aKOGREkxXqi3U,3135
|
|
23
|
+
hud/adapters/operator/tests/__init__.py,sha256=yTsDVusVXZBQL6DnXpLgKQCBRuOYUAVQ8Blk_k5GETk,41
|
|
24
|
+
hud/adapters/operator/tests/test_adapter.py,sha256=4RAXwyxAtkh-1Mlt1zJayRkcv3LWaPNEhDVTpwOZd4A,12942
|
|
25
|
+
hud/agent/__init__.py,sha256=_OxMG3UW1vXSuixdpo09b1jexfWcUbfK44zto8t6_LE,453
|
|
26
|
+
hud/agent/base.py,sha256=d7eMoRPepVSCFUyU1oV1hGvyff8rsPtXDelVcJlEF7Y,4022
|
|
27
|
+
hud/agent/claude.py,sha256=bXgdzlZHaIHaxrGIYt6w1kEh5oHFDT5P2u304swi8wU,7529
|
|
28
|
+
hud/agent/claude_plays_pokemon.py,sha256=H1AIXb6qNE-45G5CdBgsDGTMXOgjg44wGtySgeKgeq4,10009
|
|
29
|
+
hud/agent/langchain.py,sha256=iuMpu-k55Qic0LCzfOR0Wa5kDhIv7zHCc8et3axypus,8833
|
|
30
|
+
hud/agent/operator.py,sha256=Bji_v6NB-hUyTe1otdpEc3Hb0ZpyQbuL_iZu5irYFe4,8612
|
|
31
|
+
hud/agent/misc/__init__.py,sha256=-ftYH1T5r7fXKKra6d8jXYmUz9KOTmYwBrPJU-V3S7g,71
|
|
32
|
+
hud/agent/misc/response_agent.py,sha256=3PPsZqNAyUo2ouSV0ylGQj9fJqojfSB2roq2DadUdG0,3048
|
|
33
|
+
hud/agent/tests/__init__.py,sha256=HbAW7FvSvzzKPU5LpveZceU8XTcDkRe1Bmte3OGi2f0,29
|
|
34
|
+
hud/agent/tests/test_base.py,sha256=8RiJapLxO1JrAShIQ42xAyHAEb_m8MYTnH5PGC0qKRE,8501
|
|
35
|
+
hud/env/__init__.py,sha256=wVEesXMXM5hcNXQHt0-PN4-9RnE69DEnQENS7uJSv_Y,266
|
|
36
|
+
hud/env/client.py,sha256=brhfLkWGSuvxl3vqGMCQT-vXfj8rUbJMhE3zJg9WMDA,869
|
|
37
|
+
hud/env/docker_client.py,sha256=-nQLGeRl9GLJKEku1eIBL3RQuteIPLrecjVHxzSD_vU,10463
|
|
38
|
+
hud/env/environment.py,sha256=hSJh5KaiGuWw4IEV1IHNDVjOxyoyOxhHkavGj_5RBbQ,15154
|
|
39
|
+
hud/env/local_docker_client.py,sha256=bcgmmRJGX1--bf0-5Zlk59l2W0PvABnDn4FWZKvXjjk,9565
|
|
40
|
+
hud/env/remote_client.py,sha256=gjCzcuotvDC7GraVBBT2Tix5GKpntwtldv5PqnXx8wk,6109
|
|
41
|
+
hud/env/remote_docker_client.py,sha256=ZLqbd6IeU9BDndjwanmJN3_1CEsrCkntumGavLiPi88,9484
|
|
42
|
+
hud/evaluators/__init__.py,sha256=V5nktEAw3EDn2Y537pjia5Y1IjdLBIPrDjTs6YTCdX4,153
|
|
43
|
+
hud/evaluators/base.py,sha256=ALO9Rj-R_9HtHIHYp84bsQQD12De0XnCTwad78_T5-k,771
|
|
44
|
+
hud/evaluators/inspect.py,sha256=ZvrTXLpgibyvQ5aNXAMP4quyXISrRQHg9besDcuCx7U,692
|
|
45
|
+
hud/evaluators/judge.py,sha256=N3gEQGwVin9Ir80wWw6VtaL0xrlzitbmItaLm0he5gY,5962
|
|
46
|
+
hud/evaluators/match.py,sha256=8YVQD942myX72Jkme2JFIVlmKhFXEa3CgGTjLC8O5n4,4701
|
|
47
|
+
hud/evaluators/remote.py,sha256=kmD_XIU20KvX0NKgaEEKTTKHp0KVRa_3jUEgONh2nkY,2054
|
|
48
|
+
hud/evaluators/tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
49
|
+
hud/evaluators/tests/test_inspect.py,sha256=8dMjgQfXOJGcS8gP6TzoBbQiG_NYuRL6IobMG7euJdU,376
|
|
50
|
+
hud/evaluators/tests/test_judge.py,sha256=c1GaAeq_WpBVgBlx-gQncHrOPokzKNxlbgiC8W8hxYI,7829
|
|
51
|
+
hud/evaluators/tests/test_match.py,sha256=C04GoluyT9i41YZ65xEjN7tKHQbENbrpNhNtUd4ivmA,3919
|
|
52
|
+
hud/evaluators/tests/test_remote.py,sha256=YdJpyyuRLkYP0e3jTUkD3zobS2WHQPePn8yBZtYOIN4,3243
|
|
53
|
+
hud/server/__init__.py,sha256=IPxPCqtPLguryN-nBq78Sakypw2bRiE2iHv3SXG8YRk,139
|
|
54
|
+
hud/server/requests.py,sha256=AnFW4ELojjvfF6xjS2no6_fg4Rph2aR2hjPzYTede0Q,8841
|
|
55
|
+
hud/server/tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
56
|
+
hud/server/tests/test_requests.py,sha256=63YCbykcib5MxKxm-OgHJPLX3QC7hmgIwnWaYukVM6s,9077
|
|
57
|
+
hud/telemetry/__init__.py,sha256=ky48kuZD3Bt0vOf9FwZwkV_ka7O26Tvcxh7p1lMpsMk,582
|
|
58
|
+
hud/telemetry/_trace.py,sha256=W7S6CxwtmjNl4OZbA1SQHXsaNm072J9c-fjPjQomgOY,5135
|
|
59
|
+
hud/telemetry/context.py,sha256=PNbfrMgjeRTTg0nUKXYCflqn71I_cSjU8LXdvouUfc4,5209
|
|
60
|
+
hud/telemetry/exporter.py,sha256=l-r7mADcHpn6i9hhB407hx3HS4khfbhuwX0txJ2X0VQ,17986
|
|
61
|
+
hud/telemetry/mcp_models.py,sha256=YIArMtCVfC4NVvaEmUYs_kxDs0GQ-xtFFmB8jEGKaag,11342
|
|
62
|
+
hud/telemetry/instrumentation/__init__.py,sha256=vHmSqaJMMehgRNn6EN2SMoYDD12rSHkLeVmj7Uy1my0,88
|
|
63
|
+
hud/telemetry/instrumentation/mcp.py,sha256=xGAMdhTgM1ixHiDX7xkS9Ax1NCjK3u7pLWIbIh8WZIA,21925
|
|
64
|
+
hud/telemetry/instrumentation/registry.py,sha256=UVaSsEA693lvKYd5R3n3ve6GcAB1fwqubRwIVeZiNmo,1821
|
|
65
|
+
hud/telemetry/tests/__init__.py,sha256=QMN8OzfrBUDbQESwrwHCqXLdDwCjYWX8BJcpeLUJfqA,33
|
|
66
|
+
hud/telemetry/tests/test_context.py,sha256=BGRDlXXC_VbpD4cYl_o9gRQDDKb2ox1das_ZuX14NC8,6531
|
|
67
|
+
hud/telemetry/tests/test_trace.py,sha256=JzmjNRtHdQFPqLm7hOPastENg-hMJo9p8bbxJ77iXyc,10687
|
|
68
|
+
hud/utils/__init__.py,sha256=oSl_gGoS272X2VFnBYX8hLxcP2xgGoBYQXAuLhtQgw8,260
|
|
69
|
+
hud/utils/common.py,sha256=R83ntEtKr8KXG1mKcy0I_OllFHkDrPMysPhW12uBckc,3926
|
|
70
|
+
hud/utils/config.py,sha256=kGVuR7oKXjmJfeUAnbt53GesCJf_HPvvV1R6s9xjvq4,3549
|
|
71
|
+
hud/utils/misc.py,sha256=CfOv_ftLty1iEo3Rxyz4AD4nmaBkhCJVO_W-FlcyDgI,1481
|
|
72
|
+
hud/utils/progress.py,sha256=suikwFM8sdSfkV10nAOEaInDhG4XKgOSvFePg4jSj1A,5927
|
|
73
|
+
hud/utils/telemetry.py,sha256=hrVIx2rUjSGyy9IVxTZ_3Jii83PiHjyFRd5ls2whimM,1863
|
|
74
|
+
hud/utils/tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
75
|
+
hud/utils/tests/test_common.py,sha256=gbYpQKBNdbCcEH0v1UZpxLt_NW2T5sETMIJKvy8S8pY,1658
|
|
76
|
+
hud/utils/tests/test_config.py,sha256=dPlXYWuMrxX-NOYbf0vdJ27TJpfacKG8eiKOSGOcfDU,4079
|
|
77
|
+
hud/utils/tests/test_progress.py,sha256=QunwDgi_heQXhDgmC25zgjr-sFUu5FdJ_1aYigMKeIc,6351
|
|
78
|
+
hud/utils/tests/test_telemetry.py,sha256=t0An1RTBaE0dZVEpF4uwuq5k1R-PXFR5k4u71h60tx8,1224
|
|
79
|
+
hud/utils/tests/test_version.py,sha256=wDkcJbOW78FEZpIGKXST347GOm3o_NBVONZ3RhkYIM4,159
|
|
80
|
+
hud/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
81
|
+
hud_python-0.2.6.dist-info/METADATA,sha256=xXaqxhBWDKs-vkGiCi19m4wBSbbuehk4fEd1zP5Ufbg,9469
|
|
82
|
+
hud_python-0.2.6.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
|
|
83
|
+
hud_python-0.2.6.dist-info/licenses/LICENSE,sha256=yIzBheVUf86FC1bztAcr7RYWWNxyd3B-UJQ3uddg1HA,1078
|
|
84
|
+
hud_python-0.2.6.dist-info/RECORD,,
|
|
@@ -1,62 +0,0 @@
|
|
|
1
|
-
hud/__init__.py,sha256=cEmr9zVXS5upV_qcVePvApSjhPj0CmVjXolOaY3Ojuw,529
|
|
2
|
-
hud/exceptions.py,sha256=DNfaMmFGGAZsLwSVl8HILyQE5Eg5ygPLhBa7HLYcxU8,5167
|
|
3
|
-
hud/gym.py,sha256=hM1oH9DFFmxHLO9lIU4O-zkvRDCPZ5v-4BnfzNSE75E,3603
|
|
4
|
-
hud/job.py,sha256=RxXaFJQvcYbPCMIsuQRdrEyi_q3WMgdcNXypLmZqlbU,23351
|
|
5
|
-
hud/settings.py,sha256=1lOg2MieyBlmGmXiaiMk0mQvMj4N1-jcNeHCFLK5ZJA,1339
|
|
6
|
-
hud/task.py,sha256=2E4_BTqbfkertrPDkBM4M1CPYQ8Db-Iulv22f6dInWg,5112
|
|
7
|
-
hud/taskset.py,sha256=MgY9PyxKt4vO8775Qory04NA4o-BYTjirxBntslGPmk,3253
|
|
8
|
-
hud/trajectory.py,sha256=OrcRbxK_ejFp1VhJCjZnM1WCmCXxEOK4CxNjCngcsjo,3721
|
|
9
|
-
hud/types.py,sha256=_6eqLCOyxceMX_Oo8YwJQARHLb0N9zsEtc8ZCRJbf9I,2473
|
|
10
|
-
hud/adapters/__init__.py,sha256=zz24KdC_e9TJPgWo6y57_8SzevEE5ak4Cm6tXzMxwRk,266
|
|
11
|
-
hud/adapters/claude/__init__.py,sha256=i7QEF-29FLb9qxp1eYtXs-adIk_tG54tL-9g6d3xodk,100
|
|
12
|
-
hud/adapters/claude/adapter.py,sha256=xN8IHQH-xrIXxfWxoPV0VBHFvmsJZwq8PC4Ri_FU1Ew,5867
|
|
13
|
-
hud/adapters/common/__init__.py,sha256=BjdZWJVs_AKtpFrt-tNsdQRjnz7D97DFEQirJ-r0mp8,118
|
|
14
|
-
hud/adapters/common/adapter.py,sha256=fxdz1S6uwtHczLoDzmbuT0icQ49ywGMTiuVv8oJL4g4,5822
|
|
15
|
-
hud/adapters/common/types.py,sha256=k7Ic9YhgSWg3zZAtrFaOXINg06qbD1pYNzKs1obMFQE,5018
|
|
16
|
-
hud/adapters/common/tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
17
|
-
hud/adapters/common/tests/test_adapter.py,sha256=rTD36LjvytHqMIyOLDyrn0RLIkd20s6f6dwoBEarJaw,8744
|
|
18
|
-
hud/adapters/operator/__init__.py,sha256=31vTRs268_TOLd-TeQRKau5bDYy78wxCNpJFhD5_l8U,104
|
|
19
|
-
hud/adapters/operator/adapter.py,sha256=P07FsvpS11BN4IoWNJUwUPkTW6QmJoboWhxCO7exCWs,2989
|
|
20
|
-
hud/agent/__init__.py,sha256=kolJUkXlqbpULqvV4ZHaeY7ABaDKvyOOcnB5aqGtIgw,321
|
|
21
|
-
hud/agent/base.py,sha256=AfysqGBhmOt1a24AXxrpq1TVI5UHxxv1oJbUd1-hKKo,3696
|
|
22
|
-
hud/agent/claude.py,sha256=Gz5ydfdQZFPM1e1D0qfprkNzJooAXzlXYXnrug0AIFk,7428
|
|
23
|
-
hud/agent/langchain.py,sha256=giAmw2w84nugToKm-5FMlpJSB7fJh4pa_e7hERTX1Qs,8720
|
|
24
|
-
hud/agent/operator.py,sha256=DPDN8oAXjiS44brljxYsYYIQtZdodgjYTDUeDqAiFFo,8156
|
|
25
|
-
hud/env/__init__.py,sha256=wVEesXMXM5hcNXQHt0-PN4-9RnE69DEnQENS7uJSv_Y,266
|
|
26
|
-
hud/env/client.py,sha256=brhfLkWGSuvxl3vqGMCQT-vXfj8rUbJMhE3zJg9WMDA,869
|
|
27
|
-
hud/env/docker_client.py,sha256=9ltQyz30u_iitbMK2kfzs1yhs9aeC4V65Y7A1lXl23A,10200
|
|
28
|
-
hud/env/environment.py,sha256=HWnrUUwjgqOLm7xRqH0MKeGw-3mZa_ARCEnolG6xvpY,14800
|
|
29
|
-
hud/env/local_docker_client.py,sha256=OqBnlQEV4t1TlhaU-maL3noBB8YNWWQnj-iWXPyrWYc,7751
|
|
30
|
-
hud/env/remote_client.py,sha256=080Yi2na4t5bXMgtRaRQNCmuY7dK898FyjQG6vv-0WE,6156
|
|
31
|
-
hud/env/remote_docker_client.py,sha256=5akBq7g-8I0DL0nsQHW9sUAbau6PyNyDNb77ygc_-r4,7183
|
|
32
|
-
hud/evaluators/__init__.py,sha256=V5nktEAw3EDn2Y537pjia5Y1IjdLBIPrDjTs6YTCdX4,153
|
|
33
|
-
hud/evaluators/base.py,sha256=ALO9Rj-R_9HtHIHYp84bsQQD12De0XnCTwad78_T5-k,771
|
|
34
|
-
hud/evaluators/inspect.py,sha256=ZvrTXLpgibyvQ5aNXAMP4quyXISrRQHg9besDcuCx7U,692
|
|
35
|
-
hud/evaluators/judge.py,sha256=N3gEQGwVin9Ir80wWw6VtaL0xrlzitbmItaLm0he5gY,5962
|
|
36
|
-
hud/evaluators/match.py,sha256=8YVQD942myX72Jkme2JFIVlmKhFXEa3CgGTjLC8O5n4,4701
|
|
37
|
-
hud/evaluators/remote.py,sha256=kmD_XIU20KvX0NKgaEEKTTKHp0KVRa_3jUEgONh2nkY,2054
|
|
38
|
-
hud/evaluators/tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
39
|
-
hud/evaluators/tests/test_inspect.py,sha256=8dMjgQfXOJGcS8gP6TzoBbQiG_NYuRL6IobMG7euJdU,376
|
|
40
|
-
hud/evaluators/tests/test_judge.py,sha256=c1GaAeq_WpBVgBlx-gQncHrOPokzKNxlbgiC8W8hxYI,7829
|
|
41
|
-
hud/evaluators/tests/test_match.py,sha256=C04GoluyT9i41YZ65xEjN7tKHQbENbrpNhNtUd4ivmA,3919
|
|
42
|
-
hud/evaluators/tests/test_remote.py,sha256=YdJpyyuRLkYP0e3jTUkD3zobS2WHQPePn8yBZtYOIN4,3243
|
|
43
|
-
hud/server/__init__.py,sha256=IPxPCqtPLguryN-nBq78Sakypw2bRiE2iHv3SXG8YRk,139
|
|
44
|
-
hud/server/requests.py,sha256=U0WEeY8AzsLMSICebxQyF0FEbXpKieh-3Qnisd0C6a4,7881
|
|
45
|
-
hud/server/tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
46
|
-
hud/server/tests/test_requests.py,sha256=63YCbykcib5MxKxm-OgHJPLX3QC7hmgIwnWaYukVM6s,9077
|
|
47
|
-
hud/utils/__init__.py,sha256=oSl_gGoS272X2VFnBYX8hLxcP2xgGoBYQXAuLhtQgw8,260
|
|
48
|
-
hud/utils/common.py,sha256=nAhGgcOo132p5Qg48u6WV_NVN-oiag-4To--C1SYSDU,3052
|
|
49
|
-
hud/utils/config.py,sha256=kGVuR7oKXjmJfeUAnbt53GesCJf_HPvvV1R6s9xjvq4,3549
|
|
50
|
-
hud/utils/progress.py,sha256=suikwFM8sdSfkV10nAOEaInDhG4XKgOSvFePg4jSj1A,5927
|
|
51
|
-
hud/utils/telemetry.py,sha256=hrVIx2rUjSGyy9IVxTZ_3Jii83PiHjyFRd5ls2whimM,1863
|
|
52
|
-
hud/utils/tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
53
|
-
hud/utils/tests/test_common.py,sha256=gbYpQKBNdbCcEH0v1UZpxLt_NW2T5sETMIJKvy8S8pY,1658
|
|
54
|
-
hud/utils/tests/test_config.py,sha256=dPlXYWuMrxX-NOYbf0vdJ27TJpfacKG8eiKOSGOcfDU,4079
|
|
55
|
-
hud/utils/tests/test_progress.py,sha256=QunwDgi_heQXhDgmC25zgjr-sFUu5FdJ_1aYigMKeIc,6351
|
|
56
|
-
hud/utils/tests/test_telemetry.py,sha256=t0An1RTBaE0dZVEpF4uwuq5k1R-PXFR5k4u71h60tx8,1224
|
|
57
|
-
hud/utils/tests/test_version.py,sha256=bzqe0zpgWpVjnYJR_bfGi-eikKPlSkuv959T1hKBTT8,159
|
|
58
|
-
hud/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
59
|
-
hud_python-0.2.4.dist-info/METADATA,sha256=tXUQsMPRvY2wG_2YpAAcmJpNOv9NHEOOtC4Rxgh_3SI,8108
|
|
60
|
-
hud_python-0.2.4.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
|
|
61
|
-
hud_python-0.2.4.dist-info/licenses/LICENSE,sha256=yIzBheVUf86FC1bztAcr7RYWWNxyd3B-UJQ3uddg1HA,1078
|
|
62
|
-
hud_python-0.2.4.dist-info/RECORD,,
|
|
File without changes
|
|
File without changes
|