danelcsb/daniel-lfm2-350m

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.35BQuant:BF16Context Size:32kPublished:Jul 12, 2026License:otherArchitecture:Transformer Featherless Exclusive Cold

The danelcsb/daniel-lfm2-350m is a personalized LFM2-350M checkpoint developed by danelcsb, adapted with LoRA and merged for deployment. It is specifically fine-tuned for a browser-native portfolio assistant, excelling at separating Daniel-specific claims from general definitions and synthesizing information from retrieved evidence. This model is designed to handle privacy and safety refusals, make public-search tool requests when evidence is missing, and never claim to be Daniel. Its primary use is for interactive, context-aware personal assistant applications.

Loading preview...

Overview

The danelcsb/daniel-lfm2-350m is a specialized LFM2-350M checkpoint, developed by danelcsb, that has been adapted using LoRA and merged for deployment. It is specifically designed to function as a browser-native portfolio assistant for Sangbum Daniel Choi.

Key Capabilities & Training

This model was trained on 296 curated conversations, focusing on specific behavioral scopes:

  • Verified-profile answers: 177 conversations
  • Evidence-grounded definitions: 12 conversations
  • Public-retrieval decisions: 15 conversations
  • Explicitly missing profile facts: 58 conversations
  • Privacy and safety refusals: 34 conversations

The assistant is engineered to distinguish between Daniel-specific claims and general definitions. It synthesizes definitions exclusively from retrieved evidence, initiates public-search tool requests when evidence is absent, and is explicitly trained not to impersonate Daniel.

Behavioral Evaluation

In held-out behavioral evaluations, the model achieved an overall score of 84.4%, with strong performance in key areas:

  • Evidence-grounded definitions: 100.0%
  • Privacy and safety refusals: 100.0%
  • Verified-profile answers: 81.8%
  • Retrieval decisions: 75.0%
  • Missing-profile facts: 75.0%

Use Case

This model is ideal for applications requiring a highly specialized, context-aware assistant that can manage personal profile information, handle privacy boundaries, and make informed decisions based on provided context and retrieval capabilities. It learns contextual follow-up behavior and career chronology from its SFT data, making it suitable for interactive, personalized assistant roles.

Finetunes
2 models