hugoc8/zhixia-htlm-action-v2-lfm2.5-350m

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.35BQuant:BF16Context Size:32kPublished:Oct 5, 2026License:otherArchitecture:Transformer Featherless Exclusive Cold

The hugoc8/zhixia-htlm-action-v2-lfm2.5-350m is a 350 million parameter model developed by ZhiXia, fine-tuned for constrained browser-grounding action contracts. It is based on the LiquidAI/LFM2.5-350M model and specialized in generating precise JSON actions like click, type, and select based on atomic browser instructions and visible element candidates. This model serves as a decision component for browser automation, achieving high accuracy in schema validity and action matching on synthetic evaluation data. It is specifically designed for structured browser interaction rather than general knowledge or autonomous browsing.

Loading preview...

ZhiXia HTLM Action v2 — LFM2.5-350M Overview

This model, developed by ZhiXia, is a 350 million parameter (0.35B) language model specifically fine-tuned for generating precise browser actions. It combines the LiquidAI/LFM2.5-350M base model with a ZhiXia LoRA adapter, making it highly specialized for structured browser interaction tasks. It operates by taking an atomic browser instruction and a list of visible element candidates, then outputting a compact JSON action.

Key Capabilities

  • Precise Action Generation: Emits exact JSON actions for click, type, and select operations within a browser context.
  • Constrained Output: Ensures the generated select values precisely match provided options, without translation or paraphrasing.
  • High Accuracy: Achieved 100% schema validity and target-element exact match, with 98.97% full-action exact match on held-out synthetic test data.
  • Specialized Decision Component: Designed to be integrated into a larger system where a trusted host validates and executes actions, ensuring safety and policy adherence.

Good For

  • Automated Browser Interaction: Ideal for applications requiring precise, programmatic control over web elements based on specific instructions.
  • Structured UI Automation: Suitable for tasks where the model acts as a decision engine to select the correct action and target element.
  • Integration into Browser Automation Frameworks: Can be used as a core component for generating actions that are then validated and executed by a host system.

Limitations

It is crucial to note that this model is not intended for general knowledge, coding, or unconstrained autonomous actions. Its performance is evaluated on synthetic data, and it requires host-side validation and security checks before execution in a live browser environment.

Finetunes
1 models