Instructions to use alabenayed/TounsiLM-8b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use alabenayed/TounsiLM-8b with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("CohereLabs/aya-expanse-8b")
model = PeftModel.from_pretrained(base_model, "alabenayed/TounsiLM-8b")

Transformers

How to use alabenayed/TounsiLM-8b with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="alabenayed/TounsiLM-8b")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("alabenayed/TounsiLM-8b", dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use alabenayed/TounsiLM-8b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "alabenayed/TounsiLM-8b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "alabenayed/TounsiLM-8b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/alabenayed/TounsiLM-8b

SGLang

How to use alabenayed/TounsiLM-8b with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "alabenayed/TounsiLM-8b" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "alabenayed/TounsiLM-8b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "alabenayed/TounsiLM-8b" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "alabenayed/TounsiLM-8b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use alabenayed/TounsiLM-8b with Docker Model Runner:
```
docker model run hf.co/alabenayed/TounsiLM-8b
```

TounsiLM-8b / sft_train.log

alabenayed

Upload folder using huggingface_hub

4d65edc verified 11 days ago

raw

history blame contribute delete

37.3 kB

	2026-05-07 13:27:47,763 \| INFO \| Starting SFT fine-tuning job
	2026-05-07 13:27:47,763 \| INFO \| Output directory: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-07 13:27:47,763 \| INFO \| Base model: CohereLabs/aya-expanse-8b
	2026-05-07 13:27:47,763 \| INFO \| CPT adapter: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-07 13:27:47,763 \| INFO \| Dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-07 13:27:47,763 \| INFO \| Torch version: 2.11.0+cu130
	2026-05-07 13:27:47,804 \| INFO \| CUDA available: True
	2026-05-07 13:27:47,825 \| INFO \| GPU: NVIDIA GB10
	2026-05-07 13:27:47,826 \| INFO \| No Hugging Face token provided; skipping login
	2026-05-07 13:27:47,827 \| INFO \| trl version 1.3.0: assistant_only_loss is fully supported
	2026-05-07 13:27:47,827 \| INFO \| Loading configuration: batch_size=8, accumulation=4, warmup=0.03, weight_decay=0.0100
	2026-05-07 13:27:47,827 \| INFO \| Loading tokenizer from base model: CohereLabs/aya-expanse-8b
	2026-05-07 13:27:49,034 \| INFO \| Loading base model: CohereLabs/aya-expanse-8b with full bf16 precision (no quantization)
	2026-05-07 13:30:30,504 \| INFO \| Loading CPT adapter from: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-07 13:30:34,280 \| INFO \| Loading dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-07 13:30:40,321 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-07 13:30:40,717 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-07 13:30:40,818 \| INFO \| Train rows: 25340
	2026-05-07 13:30:40,819 \| INFO \| Eval rows: 6336
	2026-05-07 13:30:40,819 \| INFO \| Sample train example: [{'role': 'system', 'content': 'أنت "التيجاني"، مساعد ذكاء اصطناعي تونسي 100%. جاوب بالتونسي الدارج فقط، وبالطول المناسب للسؤال: كان يلزم قصّر، وكان يلزم فسّر أكثر. ممنوع الهلوسة أو الخروج على الموضوع.'}, {'role': 'user', 'content': 'نحس بوجيعة في العينين من كثرة غسلان الماعون بالماء السخون؟'}, {'role': 'assistant', 'content': 'البخار الصاعد من الماء السخون ينجم يتعب العين ويسبب احمرار، هذاكا علاش حاول تستعمل ماء دافي مش سخون برشة. اغسل وجهك بماء بارد بعد ما تكمل باش تبرد عينيك.'}]
	2026-05-07 13:30:40,819 \| INFO \| No checkpoint found in /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-07 13:34:03,570 \| INFO \| Starting SFT fine-tuning job
	2026-05-07 13:34:03,570 \| INFO \| Output directory: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-07 13:34:03,570 \| INFO \| Base model: CohereLabs/aya-expanse-8b
	2026-05-07 13:34:03,570 \| INFO \| CPT adapter: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-07 13:34:03,570 \| INFO \| Dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-07 13:34:03,570 \| INFO \| Torch version: 2.11.0+cu130
	2026-05-07 13:34:03,612 \| INFO \| CUDA available: True
	2026-05-07 13:34:03,631 \| INFO \| GPU: NVIDIA GB10
	2026-05-07 13:34:03,633 \| INFO \| No Hugging Face token provided; skipping login
	2026-05-07 13:34:03,633 \| INFO \| trl version 1.3.0: assistant_only_loss is fully supported
	2026-05-07 13:34:03,633 \| INFO \| Loading configuration: batch_size=8, accumulation=4, warmup=0.03, weight_decay=0.0100
	2026-05-07 13:34:03,633 \| INFO \| Loading tokenizer from base model: CohereLabs/aya-expanse-8b
	2026-05-07 13:34:04,862 \| INFO \| Loading base model: CohereLabs/aya-expanse-8b with full bf16 precision (no quantization)
	2026-05-07 13:36:22,953 \| INFO \| Loading CPT adapter from: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-07 13:36:26,760 \| INFO \| Loading dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-07 13:36:28,674 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-07 13:36:28,675 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-07 13:36:28,676 \| INFO \| Train rows: 25340
	2026-05-07 13:36:28,676 \| INFO \| Eval rows: 6336
	2026-05-07 13:36:28,677 \| INFO \| Sample train example: [{'role': 'system', 'content': 'أنت "التيجاني"، مساعد ذكاء اصطناعي تونسي 100%. جاوب بالتونسي الدارج فقط، وبالطول المناسب للسؤال: كان يلزم قصّر، وكان يلزم فسّر أكثر. ممنوع الهلوسة أو الخروج على الموضوع.'}, {'role': 'user', 'content': 'نحس بوجيعة في العينين من كثرة غسلان الماعون بالماء السخون؟'}, {'role': 'assistant', 'content': 'البخار الصاعد من الماء السخون ينجم يتعب العين ويسبب احمرار، هذاكا علاش حاول تستعمل ماء دافي مش سخون برشة. اغسل وجهك بماء بارد بعد ما تكمل باش تبرد عينيك.'}]
	2026-05-07 13:36:28,677 \| INFO \| No checkpoint found in /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-07 13:41:24,497 \| INFO \| Starting SFT fine-tuning job
	2026-05-07 13:41:24,497 \| INFO \| Output directory: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-07 13:41:24,497 \| INFO \| Base model: CohereLabs/aya-expanse-8b
	2026-05-07 13:41:24,497 \| INFO \| CPT adapter: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-07 13:41:24,497 \| INFO \| Dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-07 13:41:24,497 \| INFO \| Torch version: 2.11.0+cu130
	2026-05-07 13:41:24,548 \| INFO \| CUDA available: True
	2026-05-07 13:41:24,576 \| INFO \| GPU: NVIDIA GB10
	2026-05-07 13:41:24,577 \| INFO \| No Hugging Face token provided; skipping login
	2026-05-07 13:41:24,578 \| INFO \| trl version 1.3.0: assistant_only_loss is fully supported
	2026-05-07 13:41:24,578 \| INFO \| Loading configuration: batch_size=8, accumulation=4, warmup=0.03, weight_decay=0.0100
	2026-05-07 13:41:24,578 \| INFO \| Loading tokenizer from base model: CohereLabs/aya-expanse-8b
	2026-05-07 13:41:25,691 \| INFO \| Loading base model: CohereLabs/aya-expanse-8b with full bf16 precision (no quantization)
	2026-05-07 13:44:02,778 \| INFO \| Loading CPT adapter from: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-07 13:44:06,613 \| INFO \| Loading dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-07 13:44:08,694 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-07 13:44:08,697 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-07 13:44:08,698 \| INFO \| Train rows: 25340
	2026-05-07 13:44:08,699 \| INFO \| Eval rows: 6336
	2026-05-07 13:44:08,699 \| INFO \| Sample train example: [{'role': 'system', 'content': 'أنت "التيجاني"، مساعد ذكاء اصطناعي تونسي 100%. جاوب بالتونسي الدارج فقط، وبالطول المناسب للسؤال: كان يلزم قصّر، وكان يلزم فسّر أكثر. ممنوع الهلوسة أو الخروج على الموضوع.'}, {'role': 'user', 'content': 'نحس بوجيعة في العينين من كثرة غسلان الماعون بالماء السخون؟'}, {'role': 'assistant', 'content': 'البخار الصاعد من الماء السخون ينجم يتعب العين ويسبب احمرار، هذاكا علاش حاول تستعمل ماء دافي مش سخون برشة. اغسل وجهك بماء بارد بعد ما تكمل باش تبرد عينيك.'}]
	2026-05-07 13:44:08,699 \| INFO \| No checkpoint found in /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-07 13:44:08,752 \| INFO \| Training started
	2026-05-07 13:46:02,478 \| INFO \| Starting SFT fine-tuning job
	2026-05-07 13:46:02,479 \| INFO \| Output directory: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-07 13:46:02,479 \| INFO \| Base model: CohereLabs/aya-expanse-8b
	2026-05-07 13:46:02,479 \| INFO \| CPT adapter: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-07 13:46:02,479 \| INFO \| Dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-07 13:46:02,479 \| INFO \| Torch version: 2.11.0+cu130
	2026-05-07 13:46:02,523 \| INFO \| CUDA available: True
	2026-05-07 13:46:02,552 \| INFO \| GPU: NVIDIA GB10
	2026-05-07 13:46:02,554 \| INFO \| No Hugging Face token provided; skipping login
	2026-05-07 13:46:02,555 \| INFO \| trl version 1.3.0: assistant_only_loss is fully supported
	2026-05-07 13:46:02,555 \| INFO \| Loading configuration: batch_size=8, accumulation=4, warmup=0.03, weight_decay=0.0100
	2026-05-07 13:46:02,555 \| INFO \| Loading tokenizer from base model: CohereLabs/aya-expanse-8b
	2026-05-07 13:46:03,698 \| INFO \| Loading base model: CohereLabs/aya-expanse-8b with full bf16 precision (no quantization)
	2026-05-07 13:48:19,475 \| INFO \| Loading CPT adapter from: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-07 13:48:23,292 \| INFO \| Loading dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-07 13:48:25,199 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-07 13:48:25,202 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-07 13:48:25,203 \| INFO \| Train rows: 25340
	2026-05-07 13:48:25,203 \| INFO \| Eval rows: 6336
	2026-05-07 13:48:25,204 \| INFO \| Sample train example: [{'role': 'system', 'content': 'أنت "التيجاني"، مساعد ذكاء اصطناعي تونسي 100%. جاوب بالتونسي الدارج فقط، وبالطول المناسب للسؤال: كان يلزم قصّر، وكان يلزم فسّر أكثر. ممنوع الهلوسة أو الخروج على الموضوع.'}, {'role': 'user', 'content': 'نحس بوجيعة في العينين من كثرة غسلان الماعون بالماء السخون؟'}, {'role': 'assistant', 'content': 'البخار الصاعد من الماء السخون ينجم يتعب العين ويسبب احمرار، هذاكا علاش حاول تستعمل ماء دافي مش سخون برشة. اغسل وجهك بماء بارد بعد ما تكمل باش تبرد عينيك.'}]
	2026-05-07 13:48:25,204 \| INFO \| No checkpoint found in /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-07 13:48:25,244 \| INFO \| Training started
	2026-05-07 13:55:23,761 \| INFO \| Starting SFT fine-tuning job
	2026-05-07 13:55:23,761 \| INFO \| Output directory: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-07 13:55:23,761 \| INFO \| Base model: CohereLabs/aya-expanse-8b
	2026-05-07 13:55:23,761 \| INFO \| CPT adapter: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-07 13:55:23,761 \| INFO \| Dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-07 13:55:23,761 \| INFO \| Torch version: 2.11.0+cu130
	2026-05-07 13:55:23,803 \| INFO \| CUDA available: True
	2026-05-07 13:55:23,826 \| INFO \| GPU: NVIDIA GB10
	2026-05-07 13:55:23,828 \| INFO \| No Hugging Face token provided; skipping login
	2026-05-07 13:55:23,829 \| INFO \| trl version 1.3.0: assistant_only_loss is fully supported
	2026-05-07 13:55:23,829 \| INFO \| Loading configuration: batch_size=8, accumulation=4, warmup=0.03, weight_decay=0.0100
	2026-05-07 13:55:23,829 \| INFO \| Loading tokenizer from base model: CohereLabs/aya-expanse-8b
	2026-05-07 13:55:24,929 \| INFO \| Loading base model: CohereLabs/aya-expanse-8b with full bf16 precision (no quantization)
	2026-05-07 13:58:01,693 \| INFO \| Loading CPT adapter from: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-07 13:58:05,420 \| INFO \| Loading dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-07 13:58:07,816 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-07 13:58:08,954 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-07 13:58:09,429 \| INFO \| Train rows: 25340
	2026-05-07 13:58:09,430 \| INFO \| Eval rows: 6336
	2026-05-07 13:58:09,430 \| INFO \| Sample train example text: <BOS_TOKEN><\|START_OF_TURN_TOKEN\|><\|SYSTEM_TOKEN\|>أنت "التيجاني"، مساعد ذكاء اصطناعي تونسي 100%. جاوب بالتونسي الدارج فقط، وبالطول المناسب للسؤال: كان يلزم قصّر، وكان يلزم فسّر أكثر. ممنوع الهلوسة أو الخروج على الموضوع.<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|USER_TOKEN\|>نحس بوجيعة في العينين من كثرة غسلان الماعون بالماء السخون؟<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>البخار الصاعد من الماء السخون ينجم يتعب العين ويسبب احمرار، هذاكا علاش حاول تستعمل ماء دافي مش سخون برشة. اغسل وجهك بماء بارد بعد ما تكمل باش تبرد عينيك.<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>
	2026-05-07 13:58:09,430 \| INFO \| No checkpoint found in /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-07 13:59:25,638 \| INFO \| Starting SFT fine-tuning job
	2026-05-07 13:59:25,639 \| INFO \| Output directory: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-07 13:59:25,639 \| INFO \| Base model: CohereLabs/aya-expanse-8b
	2026-05-07 13:59:25,639 \| INFO \| CPT adapter: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-07 13:59:25,639 \| INFO \| Dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-07 13:59:25,639 \| INFO \| Torch version: 2.11.0+cu130
	2026-05-07 13:59:25,683 \| INFO \| CUDA available: True
	2026-05-07 13:59:25,709 \| INFO \| GPU: NVIDIA GB10
	2026-05-07 13:59:25,710 \| INFO \| No Hugging Face token provided; skipping login
	2026-05-07 13:59:25,711 \| INFO \| trl version 1.3.0: assistant_only_loss is fully supported
	2026-05-07 13:59:25,711 \| INFO \| Loading configuration: batch_size=8, accumulation=4, warmup=0.03, weight_decay=0.0100
	2026-05-07 13:59:25,711 \| INFO \| Loading tokenizer from base model: CohereLabs/aya-expanse-8b
	2026-05-07 13:59:26,819 \| INFO \| Loading base model: CohereLabs/aya-expanse-8b with full bf16 precision (no quantization)
	2026-05-07 14:02:05,419 \| INFO \| Loading CPT adapter from: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-07 14:02:09,265 \| INFO \| Loading dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-07 14:02:11,637 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-07 14:02:11,918 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-07 14:02:12,159 \| INFO \| Train rows: 25340
	2026-05-07 14:02:12,160 \| INFO \| Eval rows: 6336
	2026-05-07 14:02:12,160 \| INFO \| Sample train example text: <BOS_TOKEN><\|START_OF_TURN_TOKEN\|><\|SYSTEM_TOKEN\|>أنت "التيجاني"، مساعد ذكاء اصطناعي تونسي 100%. جاوب بالتونسي الدارج فقط، وبالطول المناسب للسؤال: كان يلزم قصّر، وكان يلزم فسّر أكثر. ممنوع الهلوسة أو الخروج على الموضوع.<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|USER_TOKEN\|>نحس بوجيعة في العينين من كثرة غسلان الماعون بالماء السخون؟<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>البخار الصاعد من الماء السخون ينجم يتعب العين ويسبب احمرار، هذاكا علاش حاول تستعمل ماء دافي مش سخون برشة. اغسل وجهك بماء بارد بعد ما تكمل باش تبرد عينيك.<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>
	2026-05-07 14:02:12,161 \| INFO \| No checkpoint found in /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-07 14:02:23,459 \| INFO \| Training started
	2026-05-07 14:08:07,243 \| INFO \| Starting SFT fine-tuning job
	2026-05-07 14:08:07,244 \| INFO \| Output directory: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-07 14:08:07,244 \| INFO \| Base model: CohereLabs/aya-expanse-8b
	2026-05-07 14:08:07,244 \| INFO \| CPT adapter: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-07 14:08:07,244 \| INFO \| Dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-07 14:08:07,244 \| INFO \| Torch version: 2.11.0+cu130
	2026-05-07 14:08:07,286 \| INFO \| CUDA available: True
	2026-05-07 14:08:07,309 \| INFO \| GPU: NVIDIA GB10
	2026-05-07 14:08:07,499 \| INFO \| Hugging Face login succeeded
	2026-05-07 14:08:07,500 \| INFO \| trl version 1.3.0: assistant_only_loss is fully supported
	2026-05-07 14:08:07,500 \| INFO \| Loading configuration: batch_size=8, accumulation=4, warmup=0.03, weight_decay=0.0100
	2026-05-07 14:08:07,500 \| INFO \| Loading tokenizer from base model: CohereLabs/aya-expanse-8b
	2026-05-07 14:08:08,618 \| INFO \| Loading base model: CohereLabs/aya-expanse-8b with full bf16 precision (no quantization)
	2026-05-07 14:10:19,279 \| INFO \| Loading CPT adapter from: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-07 14:10:23,076 \| INFO \| Loading dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-07 14:10:25,353 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-07 14:10:25,621 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-07 14:10:25,861 \| INFO \| Train rows: 25340
	2026-05-07 14:10:25,862 \| INFO \| Eval rows: 6336
	2026-05-07 14:10:25,862 \| INFO \| Sample train example text: <BOS_TOKEN><\|START_OF_TURN_TOKEN\|><\|SYSTEM_TOKEN\|>أنت "التيجاني"، مساعد ذكاء اصطناعي تونسي 100%. جاوب بالتونسي الدارج فقط، وبالطول المناسب للسؤال: كان يلزم قصّر، وكان يلزم فسّر أكثر. ممنوع الهلوسة أو الخروج على الموضوع.<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|USER_TOKEN\|>نحس بوجيعة في العينين من كثرة غسلان الماعون بالماء السخون؟<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>البخار الصاعد من الماء السخون ينجم يتعب العين ويسبب احمرار، هذاكا علاش حاول تستعمل ماء دافي مش سخون برشة. اغسل وجهك بماء بارد بعد ما تكمل باش تبرد عينيك.<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>
	2026-05-07 14:10:25,862 \| INFO \| No checkpoint found in /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-07 14:10:29,187 \| INFO \| Training started
	2026-05-08 04:09:43,113 \| INFO \| Training finished
	2026-05-08 04:09:43,113 \| INFO \| Saving model and tokenizer to /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-08 04:09:45,243 \| INFO \| Saved training metrics to /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft/training_metrics.json
	2026-05-08 04:09:45,243 \| INFO \| Running preview generation on a Tunisian prompt
	2026-05-08 04:09:48,487 \| INFO \| Preview prompt: عسلامة، شنوة تنصحني نعمل كي نكون تعبان وبرشة؟
	2026-05-08 04:09:48,487 \| INFO \| Preview output: <\|START_OF_TURN_TOKEN\|><\|SYSTEM_TOKEN\|>أنت "التيجاني"، مساعد ذكاء اصطناعي تونسي 100%. جاوب بالتونسي الدارج فقط، وبالطول المناسب للسؤال: كان يلزم قصّر، وكان يلزم فسّر أكثر. ممنوع الهلوسة أو الخروج على الموضوع.<\|START_OF_TURN_TOKEN\|><\|USER_TOKEN\|>عسلامة، شنوة تنصحني نعمل كي نكون تعبان وبرشة؟<\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>ح في في في في في في في في في في<\|START_OF_TURN_TOKEN\|>
	2026-05-22 00:54:42,946 \| INFO \| Starting SFT fine-tuning job
	2026-05-22 00:54:42,946 \| INFO \| Output directory: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-22 00:54:42,947 \| INFO \| Base model: CohereLabs/aya-expanse-8b
	2026-05-22 00:54:42,947 \| INFO \| CPT adapter: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-22 00:54:42,947 \| INFO \| Dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-22 00:54:42,947 \| INFO \| Torch version: 2.11.0+cu130
	2026-05-22 00:54:43,079 \| INFO \| CUDA available: True
	2026-05-22 00:54:43,096 \| INFO \| GPU: NVIDIA GB10
	2026-05-22 00:55:42,965 \| INFO \| Starting SFT fine-tuning job
	2026-05-22 00:55:42,965 \| INFO \| Output directory: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-22 00:55:42,965 \| INFO \| Base model: CohereLabs/aya-expanse-8b
	2026-05-22 00:55:42,965 \| INFO \| CPT adapter: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-22 00:55:42,965 \| INFO \| Dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-22 00:55:42,965 \| INFO \| Torch version: 2.11.0+cu130
	2026-05-22 00:55:43,096 \| INFO \| CUDA available: True
	2026-05-22 00:55:43,113 \| INFO \| GPU: NVIDIA GB10
	2026-05-22 00:55:48,743 \| INFO \| Starting SFT fine-tuning job
	2026-05-22 00:55:48,743 \| INFO \| Output directory: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-22 00:55:48,743 \| INFO \| Base model: CohereLabs/aya-expanse-8b
	2026-05-22 00:55:48,743 \| INFO \| CPT adapter: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-22 00:55:48,743 \| INFO \| Dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-22 00:55:48,743 \| INFO \| Torch version: 2.11.0+cu130
	2026-05-22 00:55:48,878 \| INFO \| CUDA available: True
	2026-05-22 00:55:48,895 \| INFO \| GPU: NVIDIA GB10
	2026-05-22 00:55:49,029 \| INFO \| Hugging Face login succeeded
	2026-05-22 00:55:49,029 \| INFO \| trl version 1.3.0: assistant_only_loss is fully supported
	2026-05-22 00:55:49,029 \| INFO \| Loading configuration: batch_size=8, accumulation=4, warmup=0.03, weight_decay=0.0100
	2026-05-22 00:55:49,029 \| INFO \| Loading tokenizer from base model: CohereLabs/aya-expanse-8b
	2026-05-22 00:55:49,803 \| INFO \| Loading base model: CohereLabs/aya-expanse-8b with full bf16 precision (no quantization)
	2026-05-22 00:56:57,725 \| INFO \| Loading CPT adapter from: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-22 00:56:59,883 \| INFO \| Loading dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-22 00:57:06,291 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-22 00:57:07,271 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-22 00:57:07,625 \| INFO \| Train rows: 25335
	2026-05-22 00:57:07,626 \| INFO \| Eval rows: 6334
	2026-05-22 00:57:07,626 \| INFO \| Sample train example text: <BOS_TOKEN><\|START_OF_TURN_TOKEN\|><\|SYSTEM_TOKEN\|>أنت "التيجاني"، مساعد ذكاء اصطناعي تونسي 100%. جاوب بالتونسي الدارج فقط، وبالطول المناسب للسؤال: كان يلزم قصّر، وكان يلزم فسّر أكثر. ممنوع الهلوسة أو الخروج على الموضوع.<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|USER_TOKEN\|>الـ سيرتيفيكاسيون في الـ لغات مهمة لـ خدمة في تونس؟<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>مهمة برشة للـ شركات الأجنبية والـ 'اوف-شور'. الـ (DELF) والـ (TOEIC) هوما الـ معيار اللي يقيموا بيه مستواك الحقيقي.<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>
	2026-05-22 00:57:07,626 \| INFO \| Detected checkpoint: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft/checkpoint-1584
	2026-05-22 00:57:16,510 \| INFO \| Training started
	2026-05-22 00:57:17,609 \| INFO \| Training finished
	2026-05-22 00:57:17,609 \| INFO \| Saving model and tokenizer to /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-22 00:57:18,156 \| INFO \| Saved training metrics to /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft/training_metrics.json
	2026-05-22 00:57:18,156 \| INFO \| Running preview generation on a Tunisian prompt
	2026-05-22 00:57:19,396 \| INFO \| Preview prompt: عسلامة، شنوة تنصحني نعمل كي نكون تعبان وبرشة؟
	2026-05-22 00:57:19,396 \| INFO \| Preview output: <\|START_OF_TURN_TOKEN\|><\|SYSTEM_TOKEN\|>أنت "التيجاني"، مساعد ذكاء اصطناعي تونسي 100%. جاوب بالتونسي الدارج فقط، وبالطول المناسب للسؤال: كان يلزم قصّر، وكان يلزم فسّر أكثر. ممنوع الهلوسة أو الخروج على الموضوع.<\|START_OF_TURN_TOKEN\|><\|USER_TOKEN\|>عسلامة، شنوة تنصحني نعمل كي نكون تعبان وبرشة؟<\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>ح في في في في<\|START_OF_TURN_TOKEN\|>
	2026-05-22 01:03:24,154 \| INFO \| Starting SFT fine-tuning job
	2026-05-22 01:03:24,154 \| INFO \| Output directory: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-22 01:03:24,154 \| INFO \| Base model: CohereLabs/aya-expanse-8b
	2026-05-22 01:03:24,154 \| INFO \| CPT adapter: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-22 01:03:24,154 \| INFO \| Dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-22 01:03:24,154 \| INFO \| Torch version: 2.11.0+cu130
	2026-05-22 01:03:24,287 \| INFO \| CUDA available: True
	2026-05-22 01:03:24,302 \| INFO \| GPU: NVIDIA GB10
	2026-05-22 01:03:24,436 \| INFO \| Hugging Face login succeeded
	2026-05-22 01:03:24,437 \| INFO \| trl version 1.3.0: assistant_only_loss is fully supported
	2026-05-22 01:03:24,437 \| INFO \| Loading configuration: batch_size=8, accumulation=4, warmup=0.03, weight_decay=0.0100
	2026-05-22 01:03:24,437 \| INFO \| Loading tokenizer from base model: CohereLabs/aya-expanse-8b
	2026-05-22 01:03:25,235 \| INFO \| Loading base model: CohereLabs/aya-expanse-8b with full bf16 precision (no quantization)
	2026-05-22 01:04:34,301 \| INFO \| Loading CPT adapter from: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-22 01:04:36,348 \| INFO \| Loading dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-22 01:04:38,428 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-22 01:04:38,550 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-22 01:04:38,657 \| INFO \| Train rows: 25335
	2026-05-22 01:04:38,658 \| INFO \| Eval rows: 6334
	2026-05-22 01:04:38,658 \| INFO \| Sample train example text: <BOS_TOKEN><\|START_OF_TURN_TOKEN\|><\|SYSTEM_TOKEN\|>أنت "التيجاني"، مساعد ذكاء اصطناعي تونسي 100%. جاوب بالتونسي الدارج فقط، وبالطول المناسب للسؤال: كان يلزم قصّر، وكان يلزم فسّر أكثر. ممنوع الهلوسة أو الخروج على الموضوع.<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|USER_TOKEN\|>الـ سيرتيفيكاسيون في الـ لغات مهمة لـ خدمة في تونس؟<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>مهمة برشة للـ شركات الأجنبية والـ 'اوف-شور'. الـ (DELF) والـ (TOEIC) هوما الـ معيار اللي يقيموا بيه مستواك الحقيقي.<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>
	2026-05-22 01:04:38,658 \| INFO \| force_restart enabled; checkpoint resume disabled
	2026-05-22 01:04:40,699 \| INFO \| Training started
	2026-05-22 12:34:27,241 \| INFO \| Starting SFT fine-tuning job
	2026-05-22 12:34:27,242 \| INFO \| Output directory: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-22 12:34:27,242 \| INFO \| Base model: CohereLabs/aya-expanse-8b
	2026-05-22 12:34:27,242 \| INFO \| CPT adapter: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-22 12:34:27,242 \| INFO \| Dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-22 12:34:27,242 \| INFO \| Torch version: 2.11.0+cu130
	2026-05-22 12:34:27,374 \| INFO \| CUDA available: True
	2026-05-22 12:34:27,399 \| INFO \| GPU: NVIDIA GB10
	2026-05-22 12:34:27,562 \| INFO \| Hugging Face login succeeded
	2026-05-22 12:34:27,563 \| INFO \| trl version 1.3.0: assistant_only_loss is fully supported
	2026-05-22 12:34:27,563 \| INFO \| Loading configuration: batch_size=8, accumulation=4, warmup=0.03, weight_decay=0.0100
	2026-05-22 12:34:27,563 \| INFO \| Loading tokenizer from base model: CohereLabs/aya-expanse-8b
	2026-05-22 12:34:28,414 \| INFO \| Loading base model: CohereLabs/aya-expanse-8b with full bf16 precision (no quantization)
	2026-05-22 12:35:55,107 \| INFO \| Loading CPT adapter from: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-22 12:35:57,709 \| INFO \| Loading dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-22 12:35:59,906 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-22 12:36:00,063 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-22 12:36:00,200 \| INFO \| Train rows: 25335
	2026-05-22 12:36:00,200 \| INFO \| Eval rows: 6334
	2026-05-22 12:36:00,201 \| INFO \| Sample train example text: <BOS_TOKEN><\|START_OF_TURN_TOKEN\|><\|SYSTEM_TOKEN\|>أنت "التيجاني"، مساعد ذكاء اصطناعي تونسي 100%. جاوب بالتونسي الدارج فقط، وبالطول المناسب للسؤال: كان يلزم قصّر، وكان يلزم فسّر أكثر. ممنوع الهلوسة أو الخروج على الموضوع.<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|USER_TOKEN\|>الـ سيرتيفيكاسيون في الـ لغات مهمة لـ خدمة في تونس؟<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>مهمة برشة للـ شركات الأجنبية والـ 'اوف-شور'. الـ (DELF) والـ (TOEIC) هوما الـ معيار اللي يقيموا بيه مستواك الحقيقي.<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>
	2026-05-22 12:36:00,201 \| INFO \| force_restart enabled; checkpoint resume disabled
	2026-05-22 12:36:02,401 \| INFO \| Training started
	2026-05-22 15:42:58,686 \| INFO \| Starting SFT fine-tuning job
	2026-05-22 15:42:58,686 \| INFO \| Output directory: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-22 15:42:58,687 \| INFO \| Base model: CohereLabs/aya-expanse-8b
	2026-05-22 15:42:58,687 \| INFO \| CPT adapter: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-22 15:42:58,687 \| INFO \| Dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-22 15:42:58,687 \| INFO \| Torch version: 2.11.0+cu130
	2026-05-22 15:42:58,821 \| INFO \| CUDA available: True
	2026-05-22 15:42:58,846 \| INFO \| GPU: NVIDIA GB10
	2026-05-22 15:42:59,224 \| INFO \| Hugging Face login succeeded
	2026-05-22 15:42:59,224 \| INFO \| trl version 1.3.0: assistant_only_loss is fully supported
	2026-05-22 15:42:59,224 \| INFO \| Loading configuration: batch_size=8, accumulation=4, warmup=0.03, weight_decay=0.0100
	2026-05-22 15:42:59,224 \| INFO \| Loading tokenizer from base model: CohereLabs/aya-expanse-8b
	2026-05-22 15:43:00,115 \| INFO \| Loading base model: CohereLabs/aya-expanse-8b with full bf16 precision (no quantization)
	2026-05-22 15:44:21,762 \| INFO \| Loading CPT adapter from: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-22 15:44:24,226 \| INFO \| Loading dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-22 15:44:26,472 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-22 15:44:26,643 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-22 15:44:26,784 \| INFO \| Train rows: 25335
	2026-05-22 15:44:26,784 \| INFO \| Eval rows: 6334
	2026-05-22 15:44:26,786 \| INFO \| Sample train example text: <BOS_TOKEN><\|START_OF_TURN_TOKEN\|><\|SYSTEM_TOKEN\|>أنت "التيجاني"، مساعد ذكاء اصطناعي تونسي 100%. جاوب بالتونسي الدارج فقط، وبالطول المناسب للسؤال: كان يلزم قصّر، وكان يلزم فسّر أكثر. ممنوع الهلوسة أو الخروج على الموضوع.<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|USER_TOKEN\|>الـ سيرتيفيكاسيون في الـ لغات مهمة لـ خدمة في تونس؟<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>مهمة برشة للـ شركات الأجنبية والـ 'اوف-شور'. الـ (DELF) والـ (TOEIC) هوما الـ معيار اللي يقيموا بيه مستواك الحقيقي.<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>
	2026-05-22 15:44:26,786 \| INFO \| force_restart enabled; checkpoint resume disabled
	2026-05-22 15:44:29,343 \| INFO \| Training started
	2026-05-22 15:44:50,452 \| INFO \| Starting SFT fine-tuning job
	2026-05-22 15:44:50,452 \| INFO \| Output directory: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-22 15:44:50,452 \| INFO \| Base model: CohereLabs/aya-expanse-8b
	2026-05-22 15:44:50,452 \| INFO \| CPT adapter: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-22 15:44:50,452 \| INFO \| Dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-22 15:44:50,452 \| INFO \| Torch version: 2.11.0+cu130
	2026-05-22 15:44:50,586 \| INFO \| CUDA available: True
	2026-05-22 15:44:50,602 \| INFO \| GPU: NVIDIA GB10
	2026-05-22 15:44:50,763 \| INFO \| Hugging Face login succeeded
	2026-05-22 15:44:50,764 \| INFO \| trl version 1.3.0: assistant_only_loss is fully supported
	2026-05-22 15:44:50,764 \| INFO \| Loading configuration: batch_size=8, accumulation=4, warmup=0.03, weight_decay=0.0100
	2026-05-22 15:44:50,764 \| INFO \| Loading tokenizer from base model: CohereLabs/aya-expanse-8b
	2026-05-22 15:44:53,754 \| INFO \| Loading base model: CohereLabs/aya-expanse-8b with full bf16 precision (no quantization)
	2026-05-22 15:46:13,999 \| INFO \| Loading CPT adapter from: /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-cpt-tunisian
	2026-05-22 15:46:16,123 \| INFO \| Loading dataset: Syrinesmati/tunisian-question-response-dataset
	2026-05-22 15:46:18,139 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-22 15:46:18,276 \| INFO \| Using dataset fields: question=instruction, answer=response
	2026-05-22 15:46:18,398 \| INFO \| Train rows: 25335
	2026-05-22 15:46:18,398 \| INFO \| Eval rows: 6334
	2026-05-22 15:46:18,399 \| INFO \| Sample train example text: <BOS_TOKEN><\|START_OF_TURN_TOKEN\|><\|SYSTEM_TOKEN\|>أنت "التيجاني"، مساعد ذكاء اصطناعي تونسي 100%. جاوب بالتونسي الدارج فقط، وبالطول المناسب للسؤال: كان يلزم قصّر، وكان يلزم فسّر أكثر. ممنوع الهلوسة أو الخروج على الموضوع.<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|USER_TOKEN\|>الـ سيرتيفيكاسيون في الـ لغات مهمة لـ خدمة في تونس؟<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>مهمة برشة للـ شركات الأجنبية والـ 'اوف-شور'. الـ (DELF) والـ (TOEIC) هوما الـ معيار اللي يقيموا بيه مستواك الحقيقي.<\|END_OF_TURN_TOKEN\|><\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>
	2026-05-22 15:46:18,399 \| INFO \| force_restart enabled; checkpoint resume disabled
	2026-05-22 15:46:20,378 \| INFO \| Training started
	2026-05-22 22:06:47,054 \| INFO \| Training finished
	2026-05-22 22:06:47,055 \| INFO \| Saving model and tokenizer to /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft
	2026-05-22 22:06:49,072 \| INFO \| Saved training metrics to /home/ala/TunisianDialogSystem/outputs/checkpoints/aya-expanse-8b-tunisian-sft/training_metrics.json
	2026-05-22 22:06:49,072 \| INFO \| Running preview generation on a Tunisian prompt
	2026-05-22 22:06:50,169 \| INFO \| Preview prompt: عسلامة، شنوة تنصحني نعمل كي نكون تعبان وبرشة؟
	2026-05-22 22:06:50,169 \| INFO \| Preview output: <\|START_OF_TURN_TOKEN\|><\|SYSTEM_TOKEN\|>أنت "التيجاني"، مساعد ذكاء اصطناعي تونسي 100%. جاوب بالتونسي الدارج فقط، وبالطول المناسب للسؤال: كان يلزم قصّر، وكان يلزم فسّر أكثر. ممنوع الهلوسة أو الخروج على الموضوع.<\|START_OF_TURN_TOKEN\|><\|USER_TOKEN\|>عسلامة، شنوة تنصحني نعمل كي نكون تعبان وبرشة؟<\|START_OF_TURN_TOKEN\|><\|CHATBOT_TOKEN\|>ح في في في<\|START_OF_TURN_TOKEN\|>