ermiaazarkhalili/LFM2.5-350M-SFT-Fable5-Glint

TEXT GENERATIONConcurrent Unit Cost:1Model Size:0.35BQuant:BF16Context Size:32kPublished:Jun 27, 2026License:otherArchitecture:Transformer Featherless Exclusive Cold

The ermiaazarkhalili/LFM2.5-350M-SFT-Fable5-Glint is a 354 million parameter causal language model, fine-tuned by ermiaazarkhalili using LoRA on the LiquidAI/LFM2.5-350M base model. It was supervised fine-tuned on a private dataset, ermiaazarkhalili/Fable-5-Glint-Clean, to specialize in instruction-following tasks. This model offers a 32768 token context length and is optimized for specific instruction-following behaviors derived from its training data.

Loading preview...

Model Overview

ermiaazarkhalili/LFM2.5-350M-SFT-Fable5-Glint is a 354 million parameter language model developed by ermiaazarkhalili. It is a LoRA (Low-Rank Adaptation) supervised fine-tune of the LiquidAI/LFM2.5-350M base model, utilizing the Lfm2ForCausalLM architecture. The fine-tuning was performed using Unsloth and TRL on a private dataset, ermiaazarkhalili/Fable-5-Glint-Clean, with a maximum sequence length of 4096 tokens.

Key Characteristics

  • Base Model: LiquidAI/LFM2.5-350M
  • Parameters: 354M
  • Fine-tuning Method: LoRA supervised fine-tuning (r=16, alpha=16) in 4-bit QLoRA precision.
  • Training Data: ermiaazarkhalili/Fable-5-Glint-Clean (private instruction-following dataset).
  • Context Length: Inherits the base model's context capabilities, with training configured for 4096 tokens.

Limitations

  • No Benchmarks: This checkpoint has not undergone any downstream benchmark evaluation; only training-loss observations are reported.
  • Inherited Biases: It inherits biases, knowledge cutoff, and failure modes from its base model.
  • Specialized Behavior: Fine-tuned on a single instruction-following dataset, its behavior outside this distribution is untested.
  • Merged Adapters: LoRA adapters are merged into the base weights, preventing detachment from this fine-tune.