LFM3-1.7B-BrowserAgent (GGUF Q8_0)

LiquidAI LFM3-1.7B fine-tuned to operate BrowserOS and BrowserOS neo through their real MCP tool surfaces — snapshot → act → verify loops, batched form filling, tab/session management, error recovery, and the request_human_help discipline for logins, one-time codes, captchas, and money approvals.

Training data

1,553 training rows (+173 held-out eval), fully synthetic, built from the open-source browseros-ai/BrowserOS repo (tool definitions, the repo's own neo agent skill, and docs) and the live tools/list of a running BrowserOS neo (25 tools). No personal information — the dataset was screened before training. Task families: inbox triage, unsubscribing, social posting, CRM updates, expense filing with receipt upload, flight search, PDF archiving, settings changes, history reopen, tab groups, saved skills, parallel multi-tab research, two-task sessions, next-call decisions from page snapshots, error recovery, and human-help cases. Element refs in training snapshots were randomized with distractor elements so the model reads the snapshot instead of memorizing positions.

Held-out eval after 1 epoch: val loss 0.092, token accuracy 97.9%.

Recipe

LoRA (rank 8, alpha 10) on the MLX 8-bit base via mlx-vlm SFT, lr 2e-4, batch 4, max seq 2048, 1 epoch (389 iters). Adapter fused back into the base, converted to GGUF with llama.cpp's LFM3 converter, quantized to Q8_0. An MLX build is kept locally; the GGUF is the LM Studio-loadable artifact today.

Files

• lfm3-1.7b-browseragent-604a-Q8_0.gguf — the model.

Notes

• Runs in LM Studio (llama.cpp engine) and llama.cpp.
• Pairs with BrowserOS neo's MCP server (browseros-neo) or the classic BrowserOS MCP server; the model emits tool calls as JSON blocks naming the tool and its arguments.
• Base model: LiquidAI/LFM3-1.7B (see its card for license terms).