base_model: webAI-Official/TwIL-8B
library_name: peft
license: other
license_name: webai-non-commercial-license-ver.-1.0
license_link: https://huggingface.co/webAI-Official/TwIL-8B/blob/main/LICENSE.md
datasets:
- FineEnvs/SmolDataEnvs-harbor-train
tags:
- trl
- grpo
- async-grpo
- lora
- agents
- harbor
- openenv
- opencode
TwIL-8B 路 Harbor 路 OpenCode (LoRA)
A LoRA adapter for webAI-Official/TwIL-8B, trained with reinforcement learning to work as the model behind the OpenCode coding agent on data-analysis tasks.
Each training rollout is a full OpenCode session in a remote sandbox: the agent reads the task, runs shell and Python tools against the task's data files, and answers. A verifier then grades the answer. The adapter was trained on those grades with TRL's asynchronous GRPO trainer.
In short: this checkpoint solves tasks about as often as the base model, with 32% fewer tool calls on the tasks both solve. On harder repository tasks rebuilt from real agent sessions (traj2envs), it is within noise of the base model (-5.7 卤 17.9 points).
Results
Evaluated with the OpenCode harness, one attempt per task at temperature 0.8, paired against the base model on the same tasks. The interval is 95%; tool calls are compared on the tasks both solved.
| Tasks | n | Base model | This adapter (step 100) | Difference (points) | Tool calls |
|---|---|---|---|---|---|
| 150 held-out tasks of FineEnvs/SmolDataEnvs-harbor-train | 150 | 77.3% | 76.7% | -0.7 卤 6.3 | -32% (104 tasks) |
| 40 held-out tasks of harness-hub/traj2envs-rl | 35 | 42.9% | 37.1% | -5.7 卤 17.9 | -27% (9 tasks) |
The main evaluation is the first row: 150 held-out tasks of FineEnvs/SmolDataEnvs-harbor-train (tasks 1,000 and up, easy and medium, none of them seen in training). Step 100 is the final checkpoint of the run.
Use with OpenCode
1. Serve the adapter with vLLM
vllm serve webAI-Official/TwIL-8B \
--enable-lora --max-lora-rank 32 \
--lora-modules twil-harbor-opencode=harness-hub/TwIL-8B-opencode \
--enable-auto-tool-choice --tool-call-parser qwen3_coder --reasoning-parser qwen3 --chat-template chat_template.jinja \
--max-model-len 131072
This is the configuration used in training and evaluation.
chat_template.jinja is in this repo (hf download harness-hub/TwIL-8B-opencode chat_template.jinja --local-dir .): it is the template training and evaluation served TwIL with.
2. Point OpenCode at it
Install OpenCode (training and evaluation used 1.19.5):
npm install -g opencode-ai@1.19.5
Add the server to ~/.config/opencode/opencode.json:
{
"$schema": "https://opencode.ai/config.json",
"provider": {
"local": {
"npm": "@ai-sdk/openai-compatible",
"options": { "baseURL": "http://localhost:8000/v1", "apiKey": "dummy" },
"models": { "twil-harbor-opencode": {} }
}
}
}
opencode run -m local/twil-harbor-opencode "Which region had the highest revenue growth in sales.csv?"
Use with llama.cpp
A GGUF version of this adapter is in harness-hub/TwIL-8B-opencode-GGUF. It runs on top of a GGUF of the base model:
hf download mradermacher/TwIL-8B-GGUF TwIL-8B.Q8_0.gguf --local-dir .
hf download harness-hub/TwIL-8B-opencode-GGUF TwIL-8B-harbor-opencode-lora-F16.gguf --local-dir .
hf download harness-hub/TwIL-8B-opencode chat_template.jinja --local-dir .
llama serve -m TwIL-8B.Q8_0.gguf --lora TwIL-8B-harbor-opencode-lora-F16.gguf \
--chat-template-file chat_template.jinja -c 65536 --port 8080 --alias twil-harbor-opencode
Then use the OpenCode configuration above with http://localhost:8080/v1.
The GGUF version has not been re-evaluated.
Training
| Method | AsyncGRPO (TRL, experimental), LoRA |
| Environment | Harbor tasks served by OpenEnv, rollouts in HF sandboxes, tool calls screened by safeagents |
| Harness | OpenCode |
| Tasks trained on, up to this checkpoint | 116 SmolDataEnvs tasks |
| Reward | the task verifier's correctness (falls back to reward), plus a tool-efficiency bonus of weight 0.1 |
| Steps | 100 (this checkpoint), of a 100-step run |
| Learning rate | 1e-5 |
| LoRA | r = 32, alpha = 64, all linear layers |
| Rollouts | 8 per task, temperature 0.8, up to 30 model calls and 1,200 s per rollout |
| Batch | 4 micro-batches of up to 65,536 tokens per step, rollouts up to 4 policy versions stale |
| Hardware | 4脳 NVIDIA H200 on HF Jobs: vLLM on one, one trainer per harness on each of the other three |
The training script is
examples/async_grpo_harbor in TRL.
Limitations
- Small evaluation. 150 tasks put the 95% interval on the paired pass@1 difference at 卤6.3 points.
- One task family. Training and the main evaluation are on data-analysis tasks. On the traj2envs repository tasks (second row), it does not improve on the base model.
- traj2envs. This checkpoint never reached the traj2envs tasks in training, so its traj2envs result measures transfer from SmolDataEnvs.
- One harness. The adapter was trained inside OpenCode's prompts and tools. With another agent, expect it to behave like the base model, or worse.
- Non-commercial license. TwIL-8B is released under the webAI non-commercial license, and so is this adapter.