Modelpublic

theprint/ProgramManager-v1-2B

sourceHugging Faceupdated 27d agoView on Hugging Face

0likes584downloads

Model Card

ProgramManager-v1-2B

A fine-tuned version of `unsloth/Qwen3.5-2B` trained on TechPM Combined 6.40k sharegpt data using Auto-SFT — an automated hyperparameter search and supervised fine-tuning pipeline.

The base model was adapted to follow the style and content of the TechPM Combined 6.40k sharegpt dataset. Expect improved performance on tasks similar to those represented in the training data.

Model Details

Property	Value
Base model	`unsloth/Qwen3.5-2B`
Training data	`data/TechPM-Combined-6.40k-sharegpt.json`
Fine-tuning epochs	2
Fine-tuning date	2026-08-26
Fine-tuning method	LoRA (merged to full 16-bit)

Training Hyperparameters

LoRA

Parameter	Value
`r`	`4`
`alpha`	`4`
`dropout`	`0.07`
`target_modules`	`['q_proj', 'v_proj', 'k_proj', 'o_proj']`

Training

Parameter	Value
`learning_rate`	`1e-05`
`batch_size`	`1`
`gradient_accumulation_steps`	`8`
`warmup_ratio`	`0.05`
`max_seq_length`	`512`
`quantization`	`none`

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model     = AutoModelForCausalLM.from_pretrained("theprint/ProgramManager-v1-2B")
tokenizer = AutoTokenizer.from_pretrained("theprint/ProgramManager-v1-2B")

Generated by [Auto-SFT](https://github.com/theprint/auto-sft) — automated LoRA fine-tuning with hyperparameter search.