Instructions to use tiny-random/deepseek-v4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tiny-random/deepseek-v4 with Transformers:

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

pipe = pipeline("text-generation", model="tiny-random/deepseek-v4")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("tiny-random/deepseek-v4")
model = AutoModelForCausalLM.from_pretrained("tiny-random/deepseek-v4", 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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use tiny-random/deepseek-v4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tiny-random/deepseek-v4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tiny-random/deepseek-v4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tiny-random/deepseek-v4

SGLang

How to use tiny-random/deepseek-v4 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 "tiny-random/deepseek-v4" \
    --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": "tiny-random/deepseek-v4",
		"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 "tiny-random/deepseek-v4" \
        --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": "tiny-random/deepseek-v4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use tiny-random/deepseek-v4 with Docker Model Runner:
```
docker model run hf.co/tiny-random/deepseek-v4
```

yujiepan commited on 9 days ago

Commit

30ad698

verified ·

1 Parent(s): 1fe154c

Upload folder using huggingface_hub

Browse files

Files changed (5) hide show

.meta.json +1 -1
README.md +38 -6
config.json +3 -3
generation_config.json +1 -1
model.safetensors +2 -2

.meta.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "safetensors": "0.8.0",
   "torch": "2.11.0+cu128",
-  "transformers": "5.14.0.dev0",
   "vllm": "0.24.0+cu129"
 }

 {
   "safetensors": "0.8.0",
   "torch": "2.11.0+cu128",
+  "transformers": "5.15.0.dev0",
   "vllm": "0.24.0+cu129"
 }

README.md CHANGED Viewed

@@ -12,7 +12,7 @@ Note:
 | File path | Size |
 |------|------|
-| model.safetensors | 144.0MB |
 ### Example usage:
@@ -20,7 +20,7 @@ Note:
 - vLLM
 ```bash
-# Not fully tested, please raise an issue if you find any problems.
 model_id=tiny-random/deepseek-v4
 vllm serve $model_id \
   --trust-remote-code \
@@ -35,6 +35,33 @@ vllm serve $model_id \
   --speculative-config '{"method":"mtp","num_speculative_tokens":2}'
 ```
 ### Codes to create this repo:
 <details><summary>Click to expand</summary>
@@ -64,11 +91,16 @@ config = {
     "hc_eps": 1e-06,
     "hc_mult": 4,
     "hc_sinkhorn_iters": 20,
-    "head_dim": 128,
     "hidden_act": "silu",
-    "hidden_size": 128,
     "index_head_dim": 128,
-    "index_n_heads": 4,
     "index_topk": 1024,
     "initializer_range": 0.02,
     "max_position_embeddings": 1048576,
@@ -437,5 +469,5 @@ main()
 - safetensors: 0.8.0
 - torch: 2.11.0+cu128
-- transformers: 5.14.0.dev0
 - vllm: 0.24.0+cu129

 | File path | Size |
 |------|------|
+| model.safetensors | 277.3MB |
 ### Example usage:
 - vLLM
 ```bash
+# Not fully tested, please report any issues if you find any problems.
 model_id=tiny-random/deepseek-v4
 vllm serve $model_id \
   --trust-remote-code \
   --speculative-config '{"method":"mtp","num_speculative_tokens":2}'
 ```
+- SGLang
+```bash
+# Tested on H20. Please report any issues if you find any problems.
+export NVCC_PREPEND_FLAGS=-allow-unsupported-compiler
+export NVCC_APPEND_FLAGS=-allow-unsupported-compiler
+export SGLANG_OPT_USE_TILELANG_MHC_PRE=0
+export SGLANG_OPT_USE_TILELANG_MHC_POST=0
+export SGLANG_OPT_DEEPGEMM_HC_PRENORM=0
+export SGLANG_OPT_USE_TILELANG_INDEXER=1
+model_id=tiny-random/deepseek-v4
+sglang serve \
+  --trust-remote-code \
+  --model-path $model_id \
+  --tp 2 \
+  --moe-runner-backend marlin \
+  --fp8-gemm-backend triton \
+  --speculative-algorithm EAGLE \
+  --speculative-num-steps 3 \
+  --speculative-eagle-topk 1 \
+  --speculative-num-draft-tokens 4 \
+  --mem-fraction-static 0.6 \
+  --disable-cuda-graph \
+  --disable-custom-all-reduce
+```
 ### Codes to create this repo:
 <details><summary>Click to expand</summary>
     "hc_eps": 1e-06,
     "hc_mult": 4,
     "hc_sinkhorn_iters": 20,
+    # SGLang's DSV4 KV-cache layout fixes the non-RoPE portion at 448
+    # elements; together with qk_rope_head_dim=64 this must be 512.
+    "head_dim": 512,
     "hidden_act": "silu",
+    # SGLang's Hopper MXFP4 Marlin path pads hidden_size to 256. Keeping
+    # the checkpoint at 128 leaves its per-32 scales at width 4 while the
+    # runtime allocates width 8, so make the checkpoint natively compatible.
+    "hidden_size": 256,
     "index_head_dim": 128,
+    "index_n_heads": 32,
     "index_topk": 1024,
     "initializer_range": 0.02,
     "max_position_embeddings": 1048576,
 - safetensors: 0.8.0
 - torch: 2.11.0+cu128
+- transformers: 5.15.0.dev0
 - vllm: 0.24.0+cu129

config.json CHANGED Viewed

@@ -10,11 +10,11 @@
   "hc_eps": 1e-06,
   "hc_mult": 4,
   "hc_sinkhorn_iters": 20,
-  "head_dim": 128,
   "hidden_act": "silu",
-  "hidden_size": 128,
   "index_head_dim": 128,
-  "index_n_heads": 4,
   "index_topk": 1024,
   "initializer_range": 0.02,
   "max_position_embeddings": 1048576,

   "hc_eps": 1e-06,
   "hc_mult": 4,
   "hc_sinkhorn_iters": 20,
+  "head_dim": 512,
   "hidden_act": "silu",
+  "hidden_size": 256,
   "index_head_dim": 128,
+  "index_n_heads": 32,
   "index_topk": 1024,
   "initializer_range": 0.02,
   "max_position_embeddings": 1048576,

generation_config.json CHANGED Viewed

@@ -5,5 +5,5 @@
   "eos_token_id": 1,
   "temperature": 1.0,
   "top_p": 1.0,
-  "transformers_version": "5.14.0.dev0"
 }

   "eos_token_id": 1,
   "temperature": 1.0,
   "top_p": 1.0,
+  "transformers_version": "5.15.0.dev0"
 }

model.safetensors CHANGED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4395c00c7788cd8a5ec1b341bb8932f33d49a2a9cbca2892cfed2c4c3f2e112e
-size 143969614

 version https://git-lfs.github.com/spec/v1
+oid sha256:bb9a57398e8d94236a97a8fe6dedec046971166dd8b7db108850c9466eaafec4
+size 277297456