Syn-2.6M

Summary

Task: Text-Generation
Total training time: 1.8 hours
Inputs: text
Outputs: text
Params: 2,604,210
Final Loss: 2.37
Important Benchmark Scores:
 1. ARC Easy - 32.11%
 2. BLiMP - 65.33%
 3. HellaSwag - 27.03%
Framework: PyTorch, transformers
Authors: Paul Courneya, Jonathan LY

Description

Syn is a Tiny Language Model (TLM) trained on 2.7 billion tokens of synthetic data. The name Syn is an abbreviation of synthetic, reflecting the type of data the model was trained on.

Model Details

• Architecture: Qwen3.5
• Hidden Size: 140
• Number of Layers: 9
• Intermediate Size: 392 (a 2.8x expansion)
• Number of Attention Heads: 4
• Number of KV Heads: 1
• Head Dim: 35
• Vocab Size: 3584
• Max Position Embeddings: 640
• Total Parameters: 2,604,210

Training

Training Details

• Maximum Learning Rate: 3e-3
• Minimum Learning Rate: 0
• Number of Epochs: 1
• Sequence Length: 256 (yes, we accidentally trained at 256 instead of 640)
• Batch Size: 428
• Eval Split Ratio: 0.006
• Gradient Accumulation Steps: 2
• Gradient Checkpointing: True
• Gradient Clipping: 1.0
• Torch Compile: True
• Torch Compile Mode: max-autotune-no-cudagraphs
• AdamW Betas: (0.9, 0.95)
• WSD Warmup Ratio: 0.02
• WSD Stable Ratio: 0.78
• WSD Decay Ratio: 0.20
• DType: float16

Dataset

Note: the byte counts are rounded.

Final Eval and Train Loss

• Train: 2.37
• Val: 2.357

Hardware

• GPU: NVIDIA RTX 2060 (used for training)
• CPU: AMD Ryzen 5 2600 (used for tokenization)

Benchmarks

ArithMark-2.0:

For a comparison with other small language models like this one, go here.

Generation Sample

Prompt : 'Artificial intelligence is'
------------------------------------------------------------
Generated:
 a form of artificial intelligence that involves creating an attractive and reliable source of information. This involves using advanced technology to create interactive, user-friendly platforms for users who are looking to use them in their daily lives.
## II. Why Use Applications?
To enhance the user experience, it's important to have the opportunity to learn how to write and understand your content properly. To ensure that you are using new tools, take some time to read, then let go of yourself or others through conversations with the user as they look at them. Additionally, by practicing self-assessment, users can gain more control over their own ideas.
### IV. Conclusion
In this lesson, we learned about the different types of apps used for frames, their importance, and its applications. We also discussed the practical aspects of the application and practical applications of Frameworks. By understanding these concepts, readers can apply these skills to various scenarios in their careers.

Use Cases

1. Educational work and research
2. Fine-tuning for downstream use
3. Deployment on edge devices
4. Or just for fun.

Limitations

1. Cannot chat, reason, code, or answer questions
2. Almost always unfactual
3. No long-context handling

License

Before using, distributing, selling, or modifying this software, you must read the license here.

Inference

#!/usr/bin/env python3

MODEL_DIR = "fromziro/Syn-2.6M"
TOKENIZER_PATH = MODEL_DIR

PROMPT = "Artificial intelligence is"
MAX_NEW_TOKENS = 256
TEMPERATURE = 0.7
TOP_P = 0.95
TOP_K = 30
REPETITION_PENALTY = 1.2
DO_SAMPLE = True

import torch
from pathlib import Path
from transformers import AutoModelForCausalLM, AutoTokenizer, PreTrainedTokenizerFast

device = (
 "cuda" if torch.cuda.is_available() else
 "mps" if torch.backends.mps.is_available() else
 "cpu"
)
print(f"Device : {device}")

def load_tokenizer(path_or_repo: str):
 p = Path(path_or_repo)

 if p.exists() and p.is_file() and p.suffix.lower() == ".json":
 tok = PreTrainedTokenizerFast(tokenizer_file=str(p.resolve()))
 else:
 tok = AutoTokenizer.from_pretrained(path_or_repo, use_fast=True)

 if tok.bos_token is None:
 tok.add_special_tokens({"bos_token": "<|bos|>"})
 if tok.eos_token is None:
 tok.add_special_tokens({"eos_token": "<|eos|>"})
 if tok.unk_token is None:
 tok.add_special_tokens({"unk_token": "<|unk|>"})
 if tok.pad_token is None:
 tok.pad_token = tok.eos_token if tok.eos_token is not None else "<|pad|>"

 tok.padding_side = "left"
 return tok

print("Loading tokenizer...")
tokenizer = load_tokenizer(TOKENIZER_PATH)
print(f" Vocab size : {len(tokenizer)}")
print(f" BOS : {tokenizer.bos_token!r}")
print(f" EOS : {tokenizer.eos_token!r}")
print(f" PAD : {tokenizer.pad_token!r} (id={tokenizer.pad_token_id})")

print(f"\nLoading model from {MODEL_DIR} ...")
model = AutoModelForCausalLM.from_pretrained(
 MODEL_DIR,
 torch_dtype=torch.float16 if device == "cuda" else torch.float32,
 low_cpu_mem_usage=True,
)

model.eval()
model.to(device)
model.config.use_cache = False
if hasattr(model, "generation_config") and model.generation_config is not None:
 model.generation_config.use_cache = False

total_params = sum(p.numel() for p in model.parameters())
print(f" Parameters : {total_params:,}")

def generate(
 prompt: str = PROMPT,
 max_new_tokens: int = MAX_NEW_TOKENS,
 temperature: float = TEMPERATURE,
 top_p: float = TOP_P,
 top_k: int = TOP_K,
 repetition_penalty: float = REPETITION_PENALTY,
 do_sample: bool = DO_SAMPLE,
) -> str:
 bos = tokenizer.bos_token or ""
 full_prompt = bos + prompt

 inputs = tokenizer(
 full_prompt,
 return_tensors="pt",
 add_special_tokens=False,
 ).to(device)

 inputs.pop("token_type_ids", None)

 gen_kwargs = dict(
 max_new_tokens=max_new_tokens,
 do_sample=do_sample,
 repetition_penalty=repetition_penalty,
 eos_token_id=tokenizer.eos_token_id,
 pad_token_id=tokenizer.pad_token_id,
 use_cache=False,
 )

 if do_sample:
 gen_kwargs["temperature"] = temperature
 gen_kwargs["top_p"] = top_p
 gen_kwargs["top_k"] = top_k

 with torch.inference_mode():
 output_ids = model.generate(**inputs, **gen_kwargs)

 prompt_len = inputs["input_ids"].shape[-1]
 new_ids = output_ids[0][prompt_len:]
 return tokenizer.decode(new_ids, skip_special_tokens=True)

if __name__ == "__main__":
 print(f"\nPrompt : {PROMPT!r}")
 print("-" * 60)
 output = generate(PROMPT)
 print("Generated:")
 print(output)

Copyright

Copyright (c) 2026 FromZero 
Copyright (c) 2026 Paul Courneya
Copyright (c) 2026 Jonathan LY

Citation

@misc{syn2.6m,
 title = {Syn-2.6M: A Tiny Languagte Model trained on 2.7B tokens of Synthetic Data},
 organization = [FromZero],
 authors = {Paul Courneya, Jonathon LY},
 year = {2026},
 url = {https://huggingface.co/fromziro/Syn-2.6M}
}