Skip to content

fix(examples): keep completion tokenizer in sync with model overrides #670

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Tianyuan Ding (likecake) wants to merge 2 commits into
base: main
Choose a base branch
from
Open

fix(examples): keep completion tokenizer in sync with model overrides #670

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ca162a9
fix(examples): keep completion tokenizer in sync with model overrides
likecake Oct 31, 2026
d154c7a
Remove example regression tests as requested
likecake Nov 4, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions docs/55-example-gsm8k.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,5 +59,8 @@ bash run_local.sh --api completion

In token-in/token-out mode, the agent tokenizes the prompt with the configured model tokenizer, sends prompt token IDs to the OpenAI-compatible Completions endpoint, receives response token IDs, and decodes them locally for answer evaluation.

The tokenizer follows the final `actor_rollout_ref.model.path`, including overrides passed after `--model`.
To select a tokenizer explicitly, pass `agentlightning.local.env_map.GSM8K_MODEL=path/to/tokenizer`.

`run_local.sh` starts `agl-server`, `agl-controller`, and Ray locally, and writes server/controller logs under `/tmp/`.
When the script exits, it cleans up the local server, controller, and Ray process it started.
3 changes: 3 additions & 0 deletions docs/65-example-search-r1.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,6 +71,9 @@ The default agent uses the Chat Completions API. To use the token-in/token-out C
examples/search_r1/run.sh --api-type completion
```

In completion mode, the tokenizer follows the final `actor_rollout_ref.model.path`, including overrides passed after `--model`.
To select a tokenizer explicitly, pass `agentlightning.local.env_map.SEARCH_R1_TOKENIZER_MODEL=path/to/tokenizer`.

To use different dataset files:

```bash
Expand Down
2 changes: 1 addition & 1 deletion examples/gsm8k/train_gsm8k_agent.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -124,7 +124,7 @@ def build_config(
raise ValueError(f"Unsupported GSM8K OpenAI API: {api}")
if api == "completion":
overrides["agentlightning"]["local"]["agent_class"] = "examples.gsm8k.gsm8k_agent.CompletionAgent"
overrides["agentlightning"]["local"]["env_map"]["GSM8K_MODEL"] = overrides["actor_rollout_ref"]["model"]["path"]
overrides["agentlightning"]["local"]["env_map"]["GSM8K_MODEL"] = "${actor_rollout_ref.model.path}"
if agl_base_url:
overrides["agentlightning"]["agl_base_url"] = agl_base_url
if agl_key is not None:
Expand Down
4 changes: 1 addition & 3 deletions examples/search_r1/train_search_r1_agent.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -133,9 +133,7 @@ def build_config(
overrides["agentlightning"]["local"]["agent_class"] = (
"examples.search_r1.agents.search_r1_agent:SearchR1CompletionAgent"
)
overrides["agentlightning"]["local"]["env_map"]["SEARCH_R1_TOKENIZER_MODEL"] = overrides["actor_rollout_ref"][
"model"
]["path"]
overrides["agentlightning"]["local"]["env_map"]["SEARCH_R1_TOKENIZER_MODEL"] = "${actor_rollout_ref.model.path}"
if agl_base_url:
overrides["agentlightning"]["agl_base_url"] = agl_base_url
if agl_key is not None:
Expand Down