You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Ameena-9B (Full Merged Model)

Qwen3.5-9B-Base with Tajik language continual pre-training, LoRA merged into base weights.

For the LoRA adapter only, see Tohirju/ameena-9B-lora.

Model Details

Parameter	Value
Base model	`unsloth/Qwen3.5-9B-Base` (9.4B params)
Method	Continual Pre-Training (CPT) with LoRA, merged
Training data	~370M tokens of Tajik text
Dataset	`Tohirju/Tajik_Pretrain_370M_Qwen35_9B`
Training steps	25 (of 954)
Precision	bf16 (16-bit merged weights)
GPU	NVIDIA H200 (140GB)
Final loss	1.597

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("Tohirju/ameena-9B-full", torch_dtype="auto", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("Tohirju/ameena-9B-full")

inputs = tokenizer("\u0422\u043e\u04b7\u0438\u043a\u0438\u0441\u0442\u043e\u043d \u043a\u0438\u0448\u0432\u0430\u0440\u0438 \u0437\u0435\u0431\u043e \u0430\u0441\u0442", return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=100, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

License

Apache 2.0 (same as base model)

Downloads last month: 11

Safetensors

Model size

9B params

Tensor type

BF16

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Tohirju/ameena-9B-full

Base model

Qwen/Qwen3.5-9B-Base

Finetuned

unsloth/Qwen3.5-9B-Base

Finetuned

(13)

this model