Instructions to use trl-internal-testing/tiny-DeepseekV4ForCausalLM with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use trl-internal-testing/tiny-DeepseekV4ForCausalLM with Transformers:

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

pipe = pipeline("text-generation", model="trl-internal-testing/tiny-DeepseekV4ForCausalLM")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("trl-internal-testing/tiny-DeepseekV4ForCausalLM")
model = AutoModelForCausalLM.from_pretrained("trl-internal-testing/tiny-DeepseekV4ForCausalLM")
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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use trl-internal-testing/tiny-DeepseekV4ForCausalLM with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "trl-internal-testing/tiny-DeepseekV4ForCausalLM"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "trl-internal-testing/tiny-DeepseekV4ForCausalLM",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/trl-internal-testing/tiny-DeepseekV4ForCausalLM

SGLang

How to use trl-internal-testing/tiny-DeepseekV4ForCausalLM 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 "trl-internal-testing/tiny-DeepseekV4ForCausalLM" \
    --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": "trl-internal-testing/tiny-DeepseekV4ForCausalLM",
		"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 "trl-internal-testing/tiny-DeepseekV4ForCausalLM" \
        --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": "trl-internal-testing/tiny-DeepseekV4ForCausalLM",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use trl-internal-testing/tiny-DeepseekV4ForCausalLM with Docker Model Runner:
```
docker model run hf.co/trl-internal-testing/tiny-DeepseekV4ForCausalLM
```

tiny-DeepseekV4ForCausalLM

File size: 2,104 Bytes

db6aed2
 
 
 
 
 
 
 
594d821
db6aed2
72bd0a3
 
db6aed2
 
ae78897
 
 
 
db6aed2
 
ae78897
 
db6aed2
 
 
 
594d821
db6aed2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
72bd0a3
db6aed2
 
 
72bd0a3
594d821
 
ae78897
db6aed2
 
 
 
 
 
 
 
 
 
 
 
 
ae78897
 
 
 
db6aed2
 
ae78897
 
db6aed2
 
 
 
 
 
 
 
 
 
ae78897
db6aed2

{
  "architectures": [
    "DeepseekV4ForCausalLM"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "bos_token_id": 0,
  "compress_ratios": [
    0,
    0,
    4,
    128
  ],
  "compress_rope_parameters": {
    "beta_fast": 32,
    "beta_slow": 1,
    "factor": 16,
    "original_max_position_embeddings": 65536,
    "partial_rotary_factor": 0.125,
    "rope_theta": 160000.0,
    "rope_type": "yarn",
    "type": "yarn"
  },
  "compress_rope_theta": 160000.0,
  "dtype": "bfloat16",
  "eos_token_id": 1,
  "first_k_dense_replace": 2,
  "hc_eps": 1e-06,
  "hc_mult": 4,
  "hc_sinkhorn_iters": 20,
  "head_dim": 512,
  "hidden_act": "silu",
  "hidden_size": 8,
  "index_head_dim": 128,
  "index_n_heads": 64,
  "index_topk": 512,
  "initializer_range": 0.02,
  "intermediate_size": 32,
  "kv_lora_rank": null,
  "max_position_embeddings": 1048576,
  "model_type": "deepseek_v4",
  "moe_intermediate_size": 2048,
  "n_group": null,
  "n_routed_experts": 4,
  "n_shared_experts": 1,
  "norm_topk_prob": true,
  "num_attention_heads": 4,
  "num_experts_per_tok": 2,
  "num_hash_layers": 3,
  "num_hidden_layers": 4,
  "num_key_value_heads": 1,
  "num_nextn_predict_layers": 1,
  "o_groups": 8,
  "o_lora_rank": 1024,
  "output_router_logits": false,
  "pad_token_id": null,
  "partial_rotary_factor": 0.125,
  "pretraining_tp": 1,
  "q_lora_rank": 1024,
  "qk_nope_head_dim": 448,
  "qk_rope_head_dim": 64,
  "rms_norm_eps": 1e-06,
  "rope_interleave": true,
  "rope_parameters": {
    "beta_fast": 32,
    "beta_slow": 1,
    "factor": 16,
    "original_max_position_embeddings": 65536,
    "partial_rotary_factor": 0.125,
    "rope_theta": 10000.0,
    "rope_type": "yarn",
    "type": "yarn"
  },
  "rope_theta": 10000.0,
  "routed_scaling_factor": 1.5,
  "router_aux_loss_coef": 0.001,
  "router_jitter_noise": 0.0,
  "scoring_func": "sqrtsoftplus",
  "sliding_window": 128,
  "swiglu_limit": 10.0,
  "tie_word_embeddings": false,
  "topk_group": null,
  "topk_method": "noaux_tc",
  "transformers_version": "5.7.0.dev0",
  "use_cache": true,
  "v_head_dim": null,
  "vocab_size": 129280
}