Quantization made by Richard Erkhov.

mistral-1L-tiny - GGUF

Model creator: https://huggingface.co/nilq/
Original model: https://huggingface.co/nilq/mistral-1L-tiny/

Name	Quant method	Size
mistral-1L-tiny.Q2_K.gguf	Q2_K	0.02GB
mistral-1L-tiny.IQ3_XS.gguf	IQ3_XS	0.02GB
mistral-1L-tiny.IQ3_S.gguf	IQ3_S	0.02GB
mistral-1L-tiny.Q3_K_S.gguf	Q3_K_S	0.02GB
mistral-1L-tiny.IQ3_M.gguf	IQ3_M	0.02GB
mistral-1L-tiny.Q3_K.gguf	Q3_K	0.02GB
mistral-1L-tiny.Q3_K_M.gguf	Q3_K_M	0.02GB
mistral-1L-tiny.Q3_K_L.gguf	Q3_K_L	0.02GB
mistral-1L-tiny.IQ4_XS.gguf	IQ4_XS	0.02GB
mistral-1L-tiny.Q4_0.gguf	Q4_0	0.02GB
mistral-1L-tiny.IQ4_NL.gguf	IQ4_NL	0.02GB
mistral-1L-tiny.Q4_K_S.gguf	Q4_K_S	0.02GB
mistral-1L-tiny.Q4_K.gguf	Q4_K	0.02GB
mistral-1L-tiny.Q4_K_M.gguf	Q4_K_M	0.02GB
mistral-1L-tiny.Q4_1.gguf	Q4_1	0.02GB
mistral-1L-tiny.Q5_0.gguf	Q5_0	0.03GB
mistral-1L-tiny.Q5_K_S.gguf	Q5_K_S	0.03GB
mistral-1L-tiny.Q5_K.gguf	Q5_K	0.03GB
mistral-1L-tiny.Q5_K_M.gguf	Q5_K_M	0.03GB
mistral-1L-tiny.Q5_1.gguf	Q5_1	0.03GB
mistral-1L-tiny.Q6_K.gguf	Q6_K	0.03GB
mistral-1L-tiny.Q8_0.gguf	Q8_0	0.04GB

Original model description:

tags: - generated_from_trainer datasets: - roneneldan/TinyStories metrics: - accuracy model-index: - name: mistral-1L-tiny results: - task: name: Causal Language Modeling type: text-generation dataset: name: roneneldan/TinyStories type: roneneldan/TinyStories metrics: - name: Accuracy type: accuracy value: 0.5792084706530948

mistral-1L-tiny

A tiny single-layer 35.1M parameter Mistral model, with a hidden size of 512, and an MLP intermediate size of 1024. This model is trained on the roneneldan/TinyStories dataset. It achieves the following results on the evaluation set:

Loss: 1.6868
Accuracy: 0.5792

Model description

This work is inspired by the 21M parameter one-layer GPT-Neo of the Tiny Stories paper. Results reproduced to acquire high-frequency checkpoints for further analysis.

Intended uses & limitations

Analysis of feature dynamics and emergence in real-world language models.

Training procedure

Trained for 90171 steps, corresponding to ~2 hours on a single H100.

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0006
train_batch_size: 64
eval_batch_size: 8
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
num_epochs: 3.0

Training results

Quite consistent English text generation.

Framework versions

Transformers 4.38.1
Pytorch 2.2.0+cu121
Datasets 2.17.1
Tokenizers 0.15.2