File size: 4,926 Bytes
0f2b560
da28396
 
0f2b560
 
 
 
 
 
 
 
eb57e28
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0f2b560
 
d2a2ed2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
da28396
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d2a2ed2
 
 
 
 
 
 
 
 
 
da28396
 
 
 
 
d2a2ed2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
da28396
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
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
---
base_model:
- Qwen/Qwen2.5-7B
tags:
- text-generation-inference
- transformers
- unsloth
- qwen2
license: apache-2.0
language:
- en
datasets:
- joackimagno/FILIPINO_RECIPES_2K_V2
metrics:
- bleu
- rouge
- meteor
model-index:
- name: MASID-v3
  results:
  - task:
      name: Text Generation
      type: text-generation
    dataset:
      name: joackimagno/FILIPINO_RECIPES_2K_V2
      type: joackimagno/FILIPINO_RECIPES_2K_V2
      split: test
    metrics:
    - name: BLEU-4
      type: bleu
      value: 0.07
    - name: METEOR
      type: meteor
      value: 0.35
    - name: ROUGE-L (F1)
      type: rouge
      value: 0.32
      unit: f1
      config: rougeL
---

# MASID-v3

**MASID-v3** is a fine-tuned version of **Qwen2.5-7B** trained specifically for **Filipino recipe generation**, with a focus on main dish preparation.

This model was trained on the **Filipino Recipes 2K V2 dataset**, a curated collection of ~2,000 authentic Filipino recipes.  
Unlike earlier variants that explored multi-stage fine-tuning, **MASID-v3 was trained directly from Qwen2.5-7B** using this dataset to specialize the model toward Filipino culinary knowledge.

The goal of MASID-v3 is to generate structured and culturally accurate Filipino main dish recipes, covering a wide range of traditional cooking methods and ingredient combinations.

---

## Model Details
- **Base Model**: [Qwen2.5-7B](https://huggingface.co/Qwen/Qwen2.5-7B)
- **Dataset**: Filipino Recipes 2K V2 (~2,000 samples)
- **Training Objective**: Recipe text generation (Filipino cuisine, main dishes)
- **Method**: Direct fine-tuning from Qwen2.5-7B

---

## Intended Use
- Assisting in **recipe writing**
- Exploring **Filipino food culture**
- Generating **cooking instructions** in natural language

---

## Limitations
- The model was trained on a relatively **small dataset (~2k samples)**.  
- May sometimes produce **hallucinated ingredients** or **inaccurate cooking steps**.  
- Not suitable for use as a **nutritional or food safety reference**.  
- Best used for **research, education, and creative applications**.

---

## Evaluation

| Dataset                            | Split | BLEU-4 | METEOR | ROUGE-L (F1) |
|------------------------------------|:-----:|:------:|:------:|:------------:|
| joackimagno/FILIPINO_RECIPES_2K_V2 | test  |  0.07  |  0.35  |    0.32      |


---

---

This Qwen2 model was trained **2× faster** with [Unsloth](https://github.com/unslothai/unsloth) and Hugging Face’s TRL library.

[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)

## Example Usage

```python
from typing import List
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig

# Load model and tokenizer
model_name = "joackimagno/MASID-v3"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.float16,
    device_map="auto",
)

# ==============================================================
# Alpaca-style prompt
# ==============================================================

SYSTEM_INSTRUCTION = (
    "You are a Filipino chef. Generate Filipino MAIN DISH recipes.\n"
    "Follow these output rules:\n"
    "1) Use standard stovetop or oven methods.\n"
    "2) Keep steps concise and logically ordered.\n"
    "3) Output FORMAT and ORDER must be exactly:\n"
    "   Recipe name, Prep time, Cook time, Total time, Servings,\n"
    "   Full Ingredients (numbered list), Instructions (numbered list)"
)

ALPACA_TEMPLATE = (
    "Below is an instruction that describes a task, paired with an input that "
    "provides further context. Write a response that appropriately completes the request.\n\n"
    "### Instruction:\n{}\n\n### Input:\n{}\n\n### Response:\n{}"
)

def make_model_input_from_ing(ing_names: List[str]) -> str:
    return (
        "Ingredients to use: " + ", ".join(ing_names) + ".\n"
        "Task: create a Filipino main dish recipe using these ingredients. "
        "Keep steps concise, clear, and coherent."
    )

# Example input
ing_names = ["Beef", "Potato", "Sili", "Carrot", "Sayote"]

alpaca_prompt = ALPACA_TEMPLATE.format(
    SYSTEM_INSTRUCTION,
    make_model_input_from_ing(ing_names),
    ""  # leave response empty for model to generate
)

# ==============================================================
# Run inference
# ==============================================================

inputs = tokenizer(alpaca_prompt, return_tensors="pt").to(model.device)

gen_config = GenerationConfig(
    max_new_tokens=512,
    temperature=0.7,
    top_p=0.9,
    do_sample=True,
)

outputs = model.generate(**inputs, generation_config=gen_config)

generated = tokenizer.decode(
    outputs[0][inputs["input_ids"].shape[1]:],
    skip_special_tokens=True
)

print(generated.strip())