File size: 3,220 Bytes
d74e26a
 
 
 
 
 
703f1ba
 
 
d74e26a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
703f1ba
d74e26a
 
 
703f1ba
d74e26a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
703f1ba
 
 
 
 
 
 
 
 
 
 
d74e26a
703f1ba
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d74e26a
 
 
 
 
703f1ba
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
---
library_name: transformers
license: mit
base_model: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
tags:
- generated_from_trainer
- gguf
- quantized
- inference
model-index:
- name: MyModel2
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# MyModel2

This model is a fine-tuned version of [deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B) on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 0.1089

## Model description

This is a fine-tuned model available in both **SafeTensors** and **GGUF** formats. The GGUF version allows efficient inference with tools like `llama.cpp` and `ctransformers`.

## Intended uses & limitations

This model can be used for various natural language processing tasks. However, it may have limitations based on the dataset and fine-tuning constraints.

## Training and evaluation data

More information needed

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- num_epochs: 5
- mixed_precision_training: Native AMP

### Training results

| Training Loss | Epoch  | Step | Validation Loss |
|:-------------:|:------:|:----:|:---------------:|
| 0.9498        | 0.2693 | 500  | 0.6119          |
| 0.6245        | 0.5385 | 1000 | 0.5831          |
| 0.5931        | 0.8078 | 1500 | 0.5462          |
| 0.561         | 1.0770 | 2000 | 0.5148          |
| 0.5312        | 1.3463 | 2500 | 0.4750          |
| 0.523         | 1.6155 | 3000 | 0.4421          |
| 0.5121        | 1.8848 | 3500 | 0.4096          |
| 0.4059        | 2.1540 | 4000 | 0.3263          |
| 0.3559        | 2.4233 | 4500 | 0.2780          |
| 0.3409        | 2.6925 | 5000 | 0.2367          |
| 0.3352        | 2.9618 | 5500 | 0.1973          |
| 0.1918        | 3.2310 | 6000 | 0.1652          |
| 0.1826        | 3.5003 | 6500 | 0.1507          |
| 0.1762        | 3.7695 | 7000 | 0.1360          |
| 0.168         | 4.0388 | 7500 | 0.1232          |
| 0.1186        | 4.3080 | 8000 | 0.1193          |
| 0.1227        | 4.5773 | 8500 | 0.1134          |
| 0.1273        | 4.8465 | 9000 | 0.1089          |

## Inference

This model supports inference via GGUF using `llama.cpp` or `ctransformers`.

### **Using `llama.cpp` (CLI)**
```bash
git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
make -j
./main -m first.gguf -p "Hello, how are you?"
```

### **Using `ctransformers` (Python)**
```python
from ctransformers import AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained(
    "your_username/your_model_repo",
    model_file="first.gguf",
    model_type="llama"
)

output = model("Hello, how are you?")
print(output)
```

## Framework versions

- Transformers 4.48.2
- Pytorch 2.5.1+cu124
- Datasets 3.2.0
- Tokenizers 0.21.0