mllmTeam/PhoneLM-1.5B

PhoneLM-1.5B is a 1.5 billion parameter decoder-only language model pre-trained on 1.1 trillion tokens.

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM

model_name = 'mllmTeam/PhoneLM-1.5B'
model = AutoModelForCausalLM.from_pretrained(model_name, device_map='cuda', trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained(model_name)

inp = tokenizer("Machine Learning is ", return_tensors="pt")
inp = {k: v.to('cuda') for k, v in inp.items()}
out = model.generate(**inp, 
                     max_length=256,
                     do_sample=True,
                     temperature=0.7,
                     top_p=0.7
                     )
text = tokenizer.decode(out[0], skip_special_tokens=True)
print(text)

Model Details

• Developed by: mllmTeam
• Model type: PhoneLM 1.5B models are auto-regressive language models based on the transformer decoder architecture.
• Language(s): English
• Paper: PhoneLM Technical Report
• Library: PhoneLM

Model Architecture

The model is a decoder-only transformer architecture with the following modifications:

Hidden Size	Layers	Heads	Sequence Length
2560	19	16	2048

• Position Embeddings: Rotary Position Embeddings (Su et al., 2021) applied to the first 25% of head embedding dimensions for improved throughput following Black et al. (2022). PhoneLM quantized the sin and cos values in Rotary Position Embeddings to 8-bit integers.
• Normalization: LayerNorm (Ba et al., 2016) with learned bias terms as opposed to RMSNorm (Zhang & Sennrich, 2019).
• Biases: We remove all bias terms from the feed-forward networks and multi-head self-attention layers, except for the biases of the query, key, and value projections (Bai et al., 2023).
• ReLU Activation Function: ReLU(Glorot et al., 2011) activation functions are adopted in feed-forward networks.
• Tokenizer: We use the SmolLM(Allal et al., 2024)'s tokenizer with a vocabulary size of 49,152.

Training Dataset

The training dataset PhoneLM used is comprised of a filtered mixture of open-source large-scale datasets available on the HuggingFace Hub: DCLM-baseline(Li et al., 2024), StarCoder (Li et al., 2023), OpenWebMath (Paster et al., 2023) and Dolma (Soldaini et al., 2024).

Evaluation Results

Model	HellaSwag	WinoGrande	PIQA	SciQ	BoolQ	ARC Easy	ARC Challenge	Average
PhoneLM-1.5B	66.9	63.0	77.3	88.8	65.5	69.7	39.9	67.31
Pythia-1.4B	52.0	57.2	71.1	79.2	63.2	53.9	28.3	57.84
OPT-1.3B	53.7	59.0	71.0	78.1	57.2	51.3	28.0	56.90
BLOOM-1.1B	43.0	54.9	67.2	74.6	59.1	45.4	25.6	52.83
TinyLlama-1.1B	59.1	58.9	73.0	82.3	58.6	55.7	31.0	59.80
MobileLLaMA-1.4B	56.1	59.4	73.0	81.9	56.7	55.8	30.3	59.03
MobiLlama-1B	62.2	59.3	74.8	82.8	60.3	56.4	31.7	61.07
OpenELM-1.1B	64.8	61.7	75.6	83.6	63.6	55.4	32.3	62.43
DCLM-1.4B	53.6	66.3	77.0	94.0	71.4	74.8	41.2	68.33
SmolLM-1.7B	49.6	60.9	75.8	93.2	66.0	76.4	43.5	66.49
Qwen 1.5-1.8B	60.9	60.5	74.2	89.4	66.5	59.1	34.7	63.61
Galactica-1.3B	41.0	54.4	63.8	87.7	62.0	58.6	30.5	56.86
StableLM 2-1.6B	68.8	64.1	75.1	76.9	80.0	60.3	39.2	66.34
Cerebras-GPT-1.3B	38.4	51.9	66.8	73.0	59.3	45.8	25.3	51.50
MiniCPM-1B	67.5	63.7	75.1	91.0	70.5	62.9	38.1	66.97
MiniCPM-2B	67.2	63.9	76.1	92.5	74.6	69.0	42.7	69.43
Gemma-2B	71.4	65.2	78.4	91.4	69.9	72.3	42.0	70.09
Gemma 2-2B	55.0	68.7	78.7	96.0	73.6	80.3	46.9	71.31

License

• This repository is released under the Apache-2.0 License.

Citation

@misc{yi2024phonelmanefficientcapablesmall,
      title={PhoneLM:an Efficient and Capable Small Language Model Family through Principled Pre-training}, 
      author={Rongjie Yi and Xiang Li and Weikai Xie and Zhenyan Lu and Chenghua Wang and Ao Zhou and Shangguang Wang and Xiwen Zhang and Mengwei Xu},
      year={2024},
      eprint={2411.05046},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2411.05046}, 
}