
Base model	`LiquidAI/LFM2.5-2.6B`
Training data	`ermiaazarkhalili/Fable-5-Complete-2M-Clean` (private)
Method	LoRA supervised fine-tuning via Unsloth + TRL
License	`other` (inherited from the base model)

Available quantizations

File	Size
`lfm2.5-2.6b-sft-fable5.q4_k_m.gguf`	1.67 GB
`lfm2.5-2.6b-sft-fable5.q5_k_m.gguf`	1.94 GB
`lfm2.5-2.6b-sft-fable5.q8_0.gguf`	2.87 GB

Usage

llama.cpp

bash

huggingface-cli download ermiaazarkhalili/LFM2.5-2.6B-SFT-Fable5-GGUF lfm2.5-2.6b-sft-fable5.q4_k_m.gguf --local-dir .
llama-cli -m lfm2.5-2.6b-sft-fable5.q4_k_m.gguf -p "Explain gradient checkpointing in two sentences." -n 256

Ollama

bash

echo 'FROM ./lfm2.5-2.6b-sft-fable5.q4_k_m.gguf' > Modelfile
ollama create lfm2.5-2.6b-sft-fable5-gguf -f Modelfile
ollama run lfm2.5-2.6b-sft-fable5-gguf

Training configuration

Setting	Value
LoRA rank (r)	16
LoRA alpha	16
Learning rate	0.0002
Epochs	2
Effective batch size	8 (2 x 4 grad accum)
Max sequence length	4096
Base precision	4-bit (QLoRA)

Observed training loss

Measured from our SLURM logs for this configuration. These are training-loss observations only — no downstream benchmark evaluation has been run on this model, so they should not be read as a quality claim.

SLURM job	Steps	First loss	Final loss
unlabelled	94,256	1.1420	0.9263

Limitations

No benchmark evaluation has been run on this checkpoint. The only reported numbers are training-loss observations.
Inherits the biases, knowledge cutoff and failure modes of the base model.
Fine-tuned on a single instruction-following dataset; behaviour outside that distribution is untested.
LoRA adapters were merged into the base weights, so the merged model cannot be detached from this fine-tune.

Reproducing

Trained by notebooks/fable_distillation_lfm2.5-2.6b_fable_unsloth.ipynb, executed non-interactively with papermill on a SLURM H100 partition (Unsloth + TRL, LoRA).

Card generated from the training run's own configuration and logs by scripts/generate_hub_model_card.py.