docs: aviary, verifiers, reasoning gym env integration docs #617
Conversation
Signed-off-by: Christian Munley <[email protected]>
cmunley1
requested review from
cwing-nvidia,
hwolff99 and
lbliii
and removed request for
lbliii
|
|
||
| ## Rollout Collection | ||
|
|
||
| ### Start vLLM Server |
There was a problem hiding this comment.
Can we follow the pattern where we are using a hosted model to generate rollouts like the quickstart?
echo "policy_base_url: https://api.openai.com/v1
policy_api_key: your-openai-api-key
policy_model_name: gpt-4.1-2025-04-14" > env.yaml
There was a problem hiding this comment.
Unnecessary burden to get a model and serve it with vLLM, right?
|
|
||
| ## Example Usage | ||
|
|
||
| ### GSM8K Environment |
There was a problem hiding this comment.
Then unlike my comment here on Reasoning gym https://github.com/NVIDIA-NeMo/Gym/pull/617/changes#r2800137030 we do not have the "setup steps" before running ng_run
Need one pattern and to follow it
|
|
||
| --- | ||
|
|
||
| ## Start Model Server |
There was a problem hiding this comment.
Similar comment to the other env tutorials about a hosted model raising barrier to entry and consistency with quickstart.
Also this one doesn't have the instruction to pull weights from HF
| ## Start Model Server | ||
|
|
||
| ```bash | ||
| uv add vllm |
There was a problem hiding this comment.
are we going with pip or uv? reasoning env has pip install https://github.com/NVIDIA-NeMo/Gym/pull/617/changes#diff-ada604f88b18e8dbff44f513c28f5aad984dc5e3bbbd213d4c1aadd9214350f9R64
3 participants
No description provided.