---
tags:
  - llaza
  - zip2zip
  - adaptive-tokenization
library_name: zip2zip
---

# zip2zip-phi35-core-v1

Training checkpoint from [zip2zip-core](https://github.com/epfl-dlab/zip2zip-core).
This is a **candidate** model (not production-ready).

## Training Config

| Field | Value |
|-------|-------|
| model_config | `Phi3.5-mini` |
| init_from | `microsoft/Phi-3.5-mini-instruct` |
| max_subtokens | 4 |
| max_codebook_size | 2048 |
| seq_len | 2048 |
| lr | 0.0003 |
| max_tokens | 262144000 |
| step | 8000 |
| data | `phi-1B-sft-8shards` |

## Usage

This is a **training checkpoint** (torchtitan format). To use for inference,
export to HuggingFace format first:

```bash
python scripts/zip2zip_hf/export_to_zip2zip.py \
    --ckpt_dir <local_path>/step_8000 \
    --output_dir <export_dir> \
    --base_model microsoft/Phi-3.5-mini-instruct \
    --model_config Phi3.5-mini
```