Llama-Express.1-Tiny is a 1B model based on Llama 3.2 (1B), fine-tuned on long chain-of-thought thinker datasets. This instruction-tuned, text-only model is optimized for multilingual dialogue use cases, including agentic retrieval and summarization tasks. It outperforms many of the available open-source and closed chat models.

Use with transformers

Starting with transformers >= 4.43.0 onward, you can run conversational inference using the Transformers pipeline abstraction or by leveraging the Auto classes with the generate() function.

Make sure to update your transformers installation via pip install --upgrade transformers.

python

import torch
from transformers import pipeline

model_id = "prithivMLmods/Llama-Express.1-Tiny"
pipe = pipeline(
    "text-generation",
    model=model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
messages = [
    {"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"},
    {"role": "user", "content": "Who are you?"},
]
outputs = pipe(
    messages,
    max_new_tokens=256,
)
print(outputs[0]["generated_text"][-1])

Intended Use

Multilingual Dialogue:
- Designed for high-quality, multilingual conversations, making it suitable for applications requiring natural, fluid dialogue across languages.
Agentic Retrieval:
- Optimized for retrieval-based tasks where reasoning and contextual chaining are crucial for extracting and summarizing relevant information.
Summarization Tasks:
- Effective in generating concise and accurate summaries from complex and lengthy texts, suitable for academic, professional, and casual use cases.
Instruction-Following Applications:
- Fine-tuned for tasks requiring adherence to user-provided instructions, making it ideal for automation workflows, content creation, and virtual assistant integrations.

Limitations

Monomodal Focus:
- As a text-only model, it cannot process multimodal inputs like images, audio, or videos, limiting its versatility in multimedia applications.
Context Length Constraints:
- While optimized for long chain-of-thought reasoning, extreme cases with very large contexts may still lead to degraded performance or truncation issues.
Bias and Ethics:
- The model might reflect biases present in the training datasets, potentially resulting in outputs that could be culturally insensitive or inappropriate.
Performance in Low-Resource Languages:
- While multilingual, its effectiveness may vary across languages, with possible performance drops in underrepresented or low-resource languages.
Dependency on Input Quality:
- The model's output is heavily influenced by the clarity and specificity of the input instructions. Ambiguous or vague prompts may lead to suboptimal results.
Lack of Real-Time Internet Access:
- Without real-time retrieval capabilities, it cannot provide up-to-date information or verify facts against the latest data.