---
license: gemma
language:
- en
base_model:
- google/gemma-2-27b-it
pipeline_tag: text-generation
library_name: transformers
tags:
- text-generation-inference
---
![8.png](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/vebaBsL6MsLveGCH3y1ig.png)

Blaze.1-27B-Reflection is a Gemma 2-based 27B parameter model. Gemma is a family of lightweight, state-of-the-art open models from Google, built using the same research and technology behind the Gemini models. These models are text-to-text, decoder-only large language models available in English, with open weights for both pre-trained and instruction-tuned variants. Gemma models are well-suited for a variety of text generation tasks, including question answering, summarization, and reasoning. Blaze.1-27B-Reflection is fine-tuned on self-reflection and behavioral data, using synthetic datasets for long-chain-of-thought reasoning from models such as DeepSeek and QwQ.

# **Quickstart Chat Template**

Below we share some code snippets on how to get quickly started with running the model. First, install the Transformers library with:
```sh
pip install -U transformers
```

Then, copy the snippet from the section that is relevant for your usecase.

# **Running with the `pipeline` API**

```python
import torch
from transformers import pipeline

pipe = pipeline(
    "text-generation",
    model="prithivMLmods/Blaze.1-27B-Reflection",
    model_kwargs={"torch_dtype": torch.bfloat16},
    device="cuda",  # replace with "mps" to run on a Mac device
)

messages = [
    {"role": "user", "content": "Who are you? Please, answer in pirate-speak."},
]

outputs = pipe(messages, max_new_tokens=256)
assistant_response = outputs[0]["generated_text"][-1]["content"].strip()
print(assistant_response)
# Ahoy, matey! I be Gemma, a digital scallywag, a language-slingin' parrot of the digital seas. I be here to help ye with yer wordy woes, answer yer questions, and spin ye yarns of the digital world.  So, what be yer pleasure, eh? 🦜
```

# **Running the model on a single / multi GPU**

```python
# pip install accelerate
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

tokenizer = AutoTokenizer.from_pretrained("prithivMLmods/Blaze.1-27B-Reflection")
model = AutoModelForCausalLM.from_pretrained(
    "prithivMLmods/Blaze.1-27B-Reflection",
    device_map="auto",
    torch_dtype=torch.bfloat16,
)

input_text = "Write me a poem about Machine Learning."
input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")

outputs = model.generate(**input_ids, max_new_tokens=32)
print(tokenizer.decode(outputs[0]))
```

You can ensure the correct chat template is applied by using `tokenizer.apply_chat_template` as follows:
```python
messages = [
    {"role": "user", "content": "Write me a poem about Machine Learning."},
]
input_ids = tokenizer.apply_chat_template(messages, return_tensors="pt", return_dict=True).to("cuda")

outputs = model.generate(**input_ids, max_new_tokens=256)
print(tokenizer.decode(outputs[0]))
```

<a name="precisions"></a>

# **Running the model on a GPU using different precisions**

The native weights of this model were exported in `bfloat16` precision.

You can also use `float32` if you skip the dtype, but no precision increase will occur (model weights will just be upcasted to `float32`). See examples below.

* _Upcasting to `torch.float32`_

```python
# pip install accelerate
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("prithivMLmods/Blaze.1-27B-Reflection")
model = AutoModelForCausalLM.from_pretrained(
    "prithivMLmods/Blaze.1-27B-Reflection",
    device_map="auto",
)

input_text = "Write me a poem about Machine Learning."
input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")

outputs = model.generate(**input_ids, max_new_tokens=32)
print(tokenizer.decode(outputs[0]))
```

# **Intended Use**  
Blaze.1-27B-Reflection is designed for advanced reasoning tasks that require long-chain-of-thought processing, self-reflection, and behavioral analysis. Its primary applications include:  

1. **Question Answering**: The model excels in providing detailed, step-by-step answers to complex queries.  
2. **Summarization**: It can generate concise summaries of large text inputs, maintaining key information and logical flow.  
3. **Reasoning and Decision Support**: With its fine-tuning on self-reflection data, it can assist in tasks that require thoughtful analysis, such as legal reasoning, policy development, and strategic planning.  
4. **Conversational AI**: Due to its instruction-tuned nature, it performs well in interactive dialogue systems, offering coherent and context-aware responses.  
5. **Creative Writing**: The model can be employed in generating high-quality content for creative tasks, including storytelling and content ideation.

# **Limitations**  
1. **Language and Domain Constraints**: While the model is effective in English, it may perform poorly with non-English inputs or domain-specific jargon outside its training scope.  
2. **Context Retention Issues**: In very long conversations or documents, the model may lose track of earlier context, leading to incomplete or off-topic responses.  
3. **Over-reliance on Synthetic Data**: Since Blaze.1-27B-Reflection is fine-tuned on synthetic datasets, it may exhibit biases or inconsistencies when faced with real-world, nuanced scenarios.  
4. **Circular Reasoning**: The model may occasionally enter recursive reasoning loops, generating verbose responses without reaching a clear conclusion.  
5. **Computational Demand**: As a 27B parameter model, it requires substantial computational resources for both inference and fine-tuning, which may limit its accessibility for users with limited hardware.  
6. **Hallucinations**: Like most large language models, it may confidently generate incorrect information, especially when asked about facts or events outside its training data.