Instructions to use serda-dev/Jamba2-3B-Turkish with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use serda-dev/Jamba2-3B-Turkish with Transformers:

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

pipe = pipeline("text-generation", model="serda-dev/Jamba2-3B-Turkish")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("serda-dev/Jamba2-3B-Turkish")
model = AutoModelForCausalLM.from_pretrained("serda-dev/Jamba2-3B-Turkish", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use serda-dev/Jamba2-3B-Turkish with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "serda-dev/Jamba2-3B-Turkish"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "serda-dev/Jamba2-3B-Turkish",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/serda-dev/Jamba2-3B-Turkish

SGLang

How to use serda-dev/Jamba2-3B-Turkish 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 "serda-dev/Jamba2-3B-Turkish" \
    --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": "serda-dev/Jamba2-3B-Turkish",
		"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 "serda-dev/Jamba2-3B-Turkish" \
        --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": "serda-dev/Jamba2-3B-Turkish",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use serda-dev/Jamba2-3B-Turkish with Docker Model Runner:
```
docker model run hf.co/serda-dev/Jamba2-3B-Turkish
```

Jamba2-3B-Turkish / benchmarks /20261014 /manifest.json

serda-dev

Publish verified full-weight final Turkish SFT and five-model benchmark

089da0e verified 19 days ago

Raw History Blame Contribute Delete

4.84 kB

{
  "seed": 20261014,
  "sources": [
    {
      "task": "turkish_mmlu",
      "repo": "AYueksel/TurkishMMLU",
      "revision": "fcbe5bb19af2714186fdd5752671e0748efa0024",
      "config": "All",
      "split": "test",
      "available": 900,
      "selected": 900,
      "skipped": {}
    },
    {
      "task": "belebele_tr",
      "repo": "facebook/belebele",
      "revision": "7899cdfa4e1e0d733fd77c848e2c273cb1d32be2",
      "config": "tur_Latn",
      "split": "test",
      "available": 900,
      "selected": 900,
      "skipped": {}
    },
    {
      "task": "gsm8k_tr",
      "repo": "malhajar/gsm8k-tr",
      "revision": "1046702fc5d8e14330ae90fdfd98735194ffa4ce",
      "config": "main",
      "split": "test",
      "available": 1318,
      "selected": 200,
      "skipped": {}
    },
    {
      "task": "ifeval_tr_subset",
      "repo": "bezir/IFEval_tr",
      "revision": "07e8c2cf4ae66a86fba57d389a73cbc10e865d5d",
      "config": null,
      "split": "train",
      "available": 350,
      "selected": 120,
      "skipped": {
        "Unsupported Turkish rule: combination:repeat_prompt": 15,
        "Unsupported Turkish rule: change_case:capital_word_frequency": 10,
        "Unsupported Turkish rule: keywords:letter_frequency": 19,
        "Unsupported Turkish rule: detectable_content:postscript": 18,
        "Unsupported Turkish rule: length_constraints:number_sentences": 29,
        "Unsupported Turkish rule: length_constraints:nth_paragraph_first_word": 9,
        "Unsupported Turkish rule: detectable_format:constrained_response": 12
      }
    }
  ],
  "models": [
    {
      "id": "jamba_sft",
      "repo": "serda-dev/Jamba2-3B-Turkish-SFT-20261013",
      "revision": "0b22f86c21610f620826067f573aa5ef2821e9cf",
      "subfolder": "checkpoints/final"
    },
    {
      "id": "qwen25_3b",
      "repo": "Qwen/Qwen2.5-3B-Instruct",
      "revision": "aa8e72537993ba99e69dfaafa59ed015b17504d1"
    },
    {
      "id": "qwen3_4b_2507",
      "repo": "Qwen/Qwen3-4B-Instruct-2507",
      "revision": "cdbee75f17c01a7cc42f958dc650907174af0554"
    },
    {
      "id": "jamba_cpt",
      "repo": "serda-dev/Jamba2-3B-Turkish-CPT-1B-20261007",
      "revision": "8b1a9c812791fa7930e64b9444b41519c023269e",
      "subfolder": "checkpoints/final"
    },
    {
      "id": "jamba_original",
      "repo": "ai21labs/AI21-Jamba2-3B",
      "revision": "525c6c8e1d9f5bddedfbdc1dbb0ade2df84230c9"
    }
  ],
  "case_sha256": "d68619d919432e6139baa3749558a8705f2bcc381ed1d136e9f87f8584ba6b8e",
  "scorer_sha256": {
    "instruction_following_eval/instructions.py": "60e086f5342a03ce8e18b64bbcccf86308f523c08aa826707a562150a52f3edf",
    "instruction_following_eval/instructions_util.py": "a73797261eee5bf447e279d82a2b700b1bdd3cb1193412dbab1270a85832bc6b",
    "instruction_following_eval/instructions_registry.py": "ec92d72c264f6d906978613085db262356174300370a3fffe6fefd5969ce9cfc",
    "instruction_following_eval/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
  },
  "protocol": "0-shot choice-letter likelihood; greedy generation; BF16 Transformers SDPA; common fixtures",
  "gsm8k_new_tokens": 512,
  "ifeval_new_tokens": 2048,
  "generation_batch": 16,
  "ifeval_notice": "Supported deterministic Turkish-safe subset, not the full published leaderboard metric.",
  "mmlu_notice": "Public AYueksel subset; not the complete 10K paper dataset or a reproduced 5-shot leaderboard.",
  "speed_notice": "Native BF16 Transformers on one GPU; excludes download/compile warmup; tokens and characters per second reported.",
  "generation_batch_schedule": "Initial samples batch 4; remaining samples batch 16. Same weights, BF16, greedy decoding and token budgets.",
  "gsm8k_cot8_new_tokens": 1024,
  "gsm8k_cot8_shots": 8,
  "gsm8k_cot8_source": "https://github.com/EleutherAI/lm-evaluation-harness/blob/main/lm_eval/tasks/gsm8k/gsm8k-cot.yaml",
  "jamba_baseline_quality_batch": 32,
  "jamba_baseline_batch_notice": "CPT and original quality runs use batch 32; fixed speed protocol remains batch 1 and 4.",
  "eos_fix": {
    "affected_models": [
      "jamba_cpt",
      "jamba_original"
    ],
    "upstream_generation_eos": 2,
    "model_and_tokenizer_eos": 519,
    "correction": "Union of tokenizer, model config and generation config terminal IDs",
    "scope": "Rerun 1040 generative evaluations; preserve valid MCQ results and common original GPU speed cohort"
  },
  "speed_instance_id": 55841484,
  "corrected_generation_cohort": {
    "instance_id": 55857180,
    "gpu": "RTX 4090",
    "machine_id": 47283,
    "public_ipaddr": "95.3.33.46",
    "scope": "Only original/CPT generation EOS correction; same physical GPU host as common speed cohort",
    "numeric_notice": "Same BF16 Transformers backend and hardware; batching may cause small numerical differences"
  }
}

	{
	"seed": 20261014,
	"sources": [
	{
	"task": "turkish_mmlu",
	"repo": "AYueksel/TurkishMMLU",
	"revision": "fcbe5bb19af2714186fdd5752671e0748efa0024",
	"config": "All",
	"split": "test",
	"available": 900,
	"selected": 900,
	"skipped": {}
	},
	{
	"task": "belebele_tr",
	"repo": "facebook/belebele",
	"revision": "7899cdfa4e1e0d733fd77c848e2c273cb1d32be2",
	"config": "tur_Latn",
	"split": "test",
	"available": 900,
	"selected": 900,
	"skipped": {}
	},
	{
	"task": "gsm8k_tr",
	"repo": "malhajar/gsm8k-tr",
	"revision": "1046702fc5d8e14330ae90fdfd98735194ffa4ce",
	"config": "main",
	"split": "test",
	"available": 1318,
	"selected": 200,
	"skipped": {}
	},
	{
	"task": "ifeval_tr_subset",
	"repo": "bezir/IFEval_tr",
	"revision": "07e8c2cf4ae66a86fba57d389a73cbc10e865d5d",
	"config": null,
	"split": "train",
	"available": 350,
	"selected": 120,
	"skipped": {
	"Unsupported Turkish rule: combination:repeat_prompt": 15,
	"Unsupported Turkish rule: change_case:capital_word_frequency": 10,
	"Unsupported Turkish rule: keywords:letter_frequency": 19,
	"Unsupported Turkish rule: detectable_content:postscript": 18,
	"Unsupported Turkish rule: length_constraints:number_sentences": 29,
	"Unsupported Turkish rule: length_constraints:nth_paragraph_first_word": 9,
	"Unsupported Turkish rule: detectable_format:constrained_response": 12
	}
	}
	],
	"models": [
	{
	"id": "jamba_sft",
	"repo": "serda-dev/Jamba2-3B-Turkish-SFT-20261013",
	"revision": "0b22f86c21610f620826067f573aa5ef2821e9cf",
	"subfolder": "checkpoints/final"
	},
	{
	"id": "qwen25_3b",
	"repo": "Qwen/Qwen2.5-3B-Instruct",
	"revision": "aa8e72537993ba99e69dfaafa59ed015b17504d1"
	},
	{
	"id": "qwen3_4b_2507",
	"repo": "Qwen/Qwen3-4B-Instruct-2507",
	"revision": "cdbee75f17c01a7cc42f958dc650907174af0554"
	},
	{
	"id": "jamba_cpt",
	"repo": "serda-dev/Jamba2-3B-Turkish-CPT-1B-20261007",
	"revision": "8b1a9c812791fa7930e64b9444b41519c023269e",
	"subfolder": "checkpoints/final"
	},
	{
	"id": "jamba_original",
	"repo": "ai21labs/AI21-Jamba2-3B",
	"revision": "525c6c8e1d9f5bddedfbdc1dbb0ade2df84230c9"
	}
	],
	"case_sha256": "d68619d919432e6139baa3749558a8705f2bcc381ed1d136e9f87f8584ba6b8e",
	"scorer_sha256": {
	"instruction_following_eval/instructions.py": "60e086f5342a03ce8e18b64bbcccf86308f523c08aa826707a562150a52f3edf",
	"instruction_following_eval/instructions_util.py": "a73797261eee5bf447e279d82a2b700b1bdd3cb1193412dbab1270a85832bc6b",
	"instruction_following_eval/instructions_registry.py": "ec92d72c264f6d906978613085db262356174300370a3fffe6fefd5969ce9cfc",
	"instruction_following_eval/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
	},
	"protocol": "0-shot choice-letter likelihood; greedy generation; BF16 Transformers SDPA; common fixtures",
	"gsm8k_new_tokens": 512,
	"ifeval_new_tokens": 2048,
	"generation_batch": 16,
	"ifeval_notice": "Supported deterministic Turkish-safe subset, not the full published leaderboard metric.",
	"mmlu_notice": "Public AYueksel subset; not the complete 10K paper dataset or a reproduced 5-shot leaderboard.",
	"speed_notice": "Native BF16 Transformers on one GPU; excludes download/compile warmup; tokens and characters per second reported.",
	"generation_batch_schedule": "Initial samples batch 4; remaining samples batch 16. Same weights, BF16, greedy decoding and token budgets.",
	"gsm8k_cot8_new_tokens": 1024,
	"gsm8k_cot8_shots": 8,
	"gsm8k_cot8_source": "https://github.com/EleutherAI/lm-evaluation-harness/blob/main/lm_eval/tasks/gsm8k/gsm8k-cot.yaml",
	"jamba_baseline_quality_batch": 32,
	"jamba_baseline_batch_notice": "CPT and original quality runs use batch 32; fixed speed protocol remains batch 1 and 4.",
	"eos_fix": {
	"affected_models": [
	"jamba_cpt",
	"jamba_original"
	],
	"upstream_generation_eos": 2,
	"model_and_tokenizer_eos": 519,
	"correction": "Union of tokenizer, model config and generation config terminal IDs",
	"scope": "Rerun 1040 generative evaluations; preserve valid MCQ results and common original GPU speed cohort"
	},
	"speed_instance_id": 55841484,
	"corrected_generation_cohort": {
	"instance_id": 55857180,
	"gpu": "RTX 4090",
	"machine_id": 47283,
	"public_ipaddr": "95.3.33.46",
	"scope": "Only original/CPT generation EOS correction; same physical GPU host as common speed cohort",
	"numeric_notice": "Same BF16 Transformers backend and hardware; batching may cause small numerical differences"
	}
	}