README.md

Quantization made by Richard Erkhov.

[Github](https://github.com/RichardErkhov)

[Discord](https://discord.gg/pvy7H8DZMG)

[Request more models](https://github.com/RichardErkhov/quant_request)


deepseek-coder-1.3B-kexer - GGUF
- Model creator: https://huggingface.co/JetBrains/
- Original model: https://huggingface.co/JetBrains/deepseek-coder-1.3B-kexer/


| Name | Quant method | Size |
| ---- | ---- | ---- |
| [deepseek-coder-1.3B-kexer.Q2_K.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.Q2_K.gguf) | Q2_K | 0.52GB |
| [deepseek-coder-1.3B-kexer.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.IQ3_XS.gguf) | IQ3_XS | 0.57GB |
| [deepseek-coder-1.3B-kexer.IQ3_S.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.IQ3_S.gguf) | IQ3_S | 0.6GB |
| [deepseek-coder-1.3B-kexer.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.Q3_K_S.gguf) | Q3_K_S | 0.6GB |
| [deepseek-coder-1.3B-kexer.IQ3_M.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.IQ3_M.gguf) | IQ3_M | 0.63GB |
| [deepseek-coder-1.3B-kexer.Q3_K.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.Q3_K.gguf) | Q3_K | 0.66GB |
| [deepseek-coder-1.3B-kexer.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.Q3_K_M.gguf) | Q3_K_M | 0.66GB |
| [deepseek-coder-1.3B-kexer.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.Q3_K_L.gguf) | Q3_K_L | 0.69GB |
| [deepseek-coder-1.3B-kexer.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.IQ4_XS.gguf) | IQ4_XS | 0.7GB |
| [deepseek-coder-1.3B-kexer.Q4_0.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.Q4_0.gguf) | Q4_0 | 0.72GB |
| [deepseek-coder-1.3B-kexer.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.IQ4_NL.gguf) | IQ4_NL | 0.73GB |
| [deepseek-coder-1.3B-kexer.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.Q4_K_S.gguf) | Q4_K_S | 0.76GB |
| [deepseek-coder-1.3B-kexer.Q4_K.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.Q4_K.gguf) | Q4_K | 0.81GB |
| [deepseek-coder-1.3B-kexer.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.Q4_K_M.gguf) | Q4_K_M | 0.81GB |
| [deepseek-coder-1.3B-kexer.Q4_1.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.Q4_1.gguf) | Q4_1 | 0.8GB |
| [deepseek-coder-1.3B-kexer.Q5_0.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.Q5_0.gguf) | Q5_0 | 0.87GB |
| [deepseek-coder-1.3B-kexer.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.Q5_K_S.gguf) | Q5_K_S | 0.89GB |
| [deepseek-coder-1.3B-kexer.Q5_K.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.Q5_K.gguf) | Q5_K | 0.93GB |
| [deepseek-coder-1.3B-kexer.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.Q5_K_M.gguf) | Q5_K_M | 0.93GB |
| [deepseek-coder-1.3B-kexer.Q5_1.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.Q5_1.gguf) | Q5_1 | 0.95GB |
| [deepseek-coder-1.3B-kexer.Q6_K.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.Q6_K.gguf) | Q6_K | 1.09GB |
| [deepseek-coder-1.3B-kexer.Q8_0.gguf](https://huggingface.co/RichardErkhov/JetBrains_-_deepseek-coder-1.3B-kexer-gguf/blob/main/deepseek-coder-1.3B-kexer.Q8_0.gguf) | Q8_0 | 1.33GB |




Original model description:
---
license: apache-2.0
datasets:
- JetBrains/KExercises
base_model: deepseek-ai/deepseek-coder-1.3b-base
results:
- task:
    type: text-generation
  dataset:
    name: MultiPL-HumanEval (Kotlin)
    type: openai_humaneval
  metrics:
  - name: pass@1
    type: pass@1
    value: 36.65
tags:
- code
---

# Kexer models

Kexer models are a collection of open-source generative text models fine-tuned on the [Kotlin Exercices](https://huggingface.co/datasets/JetBrains/KExercises) dataset. 
This is a repository for the fine-tuned **Deepseek-coder-1.3b** model in the *Hugging Face Transformers* format. 

# How to use

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

# Load pre-trained model and tokenizer
model_name = 'JetBrains/deepseek-coder-1.3B-kexer'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name).to('cuda')

# Create and encode input
input_text = """\
This function takes an integer n and returns factorial of a number:
fun factorial(n: Int): Int {\
"""
input_ids = tokenizer.encode(
    input_text, return_tensors='pt'
).to('cuda')

# Generate
output = model.generate(
    input_ids, max_length=60, num_return_sequences=1, 
    early_stopping=True, pad_token_id=tokenizer.eos_token_id,
)

# Decode output
generated_text = tokenizer.decode(output[0], skip_special_tokens=True)
print(generated_text)
```

As with the base model, we can use FIM. To do this, the following format must be used: 
```
'<｜fim▁begin｜>' + prefix + '<｜fim▁hole｜>' + suffix + '<｜fim▁end｜>'
```

# Training setup

The model was trained on one A100 GPU with following hyperparameters:

|         **Hyperparameter**           |             **Value**              |
|:---------------------------:|:----------------------------------------:|
|           `warmup`            |           10%            |
|        `max_lr`        |          1e-4          |
|        `scheduler`        |          linear          |
|        `total_batch_size`        |          256 (~130K tokens per step)          |
|        `num_epochs`        |          4          |

More details about fine-tuning can be found in the technical report (coming soon!).

# Fine-tuning data

For tuning this model, we used 15K exmaples from the synthetically generated [Kotlin Exercices](https://huggingface.co/datasets/JetBrains/KExercises) dataset. Every example follows the HumanEval format. In total, the dataset contains about 3.5M tokens. 

# Evaluation 

For evaluation, we used the [Kotlin HumanEval](https://huggingface.co/datasets/JetBrains/Kotlin_HumanEval) dataset, which contains all 161 tasks from HumanEval translated into Kotlin by human experts. You can find more details about the pre-processing necessary to obtain our results, including the code for running, on the [datasets's page](https://huggingface.co/datasets/JetBrains/Kotlin_HumanEval).

Here are the results of our evaluation:

|         **Model name**           |             **Kotlin HumanEval Pass Rate**              |
|:---------------------------:|:----------------------------------------:|
|           `Deepseek-coder-1.3B`            |           26.71            |
|        `Deepseek-coder-1.3B-Kexer`        |          **36.65**         |

# Ethical considerations and limitations

Deepseek-coder-1.3B-Kexer is a new technology that carries risks with use. The testing conducted to date has not covered, nor could it cover all scenarios. For these reasons, as with all LLMs, Deepseek-coder-1.3B-Kexer's potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate or objectionable responses to user prompts. The model was fine-tuned on a specific data format (Kotlin tasks), and deviation from this format can also lead to inaccurate or undesirable responses to user queries. Therefore, before deploying any applications of Deepseek-coder-1.3B-Kexer, developers should perform safety testing and tuning tailored to their specific applications of the model.