Text Generation
GGUF
Safetensors
English
deepseek2
llama.cpp
mixture-of-experts
multi-head-latent-attention
iq4_nl
mark-ai
systems-engineering
custom_code
conversational
Instructions to use miceworks/Mark-SLM-V3-150M with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use miceworks/Mark-SLM-V3-150M with llama.cpp:
Install (macOS, Linux)
curl -LsSf https://llama.app/install.sh | sh # Start a local OpenAI-compatible server with a web UI: llama serve -hf miceworks/Mark-SLM-V3-150M:IQ4_NL # Run inference directly in the terminal: llama cli -hf miceworks/Mark-SLM-V3-150M:IQ4_NL
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf miceworks/Mark-SLM-V3-150M:IQ4_NL # Run inference directly in the terminal: llama cli -hf miceworks/Mark-SLM-V3-150M:IQ4_NL
Build from source code
git clone https://github.com/ggml-org/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-app # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama serve -hf miceworks/Mark-SLM-V3-150M:IQ4_NL # Run inference directly in the terminal: ./build/bin/llama cli -hf miceworks/Mark-SLM-V3-150M:IQ4_NL
Use Docker
docker model run hf.co/miceworks/Mark-SLM-V3-150M:IQ4_NL
- Llama new
- LM Studio
- Jan
- vLLM
How to use miceworks/Mark-SLM-V3-150M with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "miceworks/Mark-SLM-V3-150M" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "miceworks/Mark-SLM-V3-150M", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/miceworks/Mark-SLM-V3-150M:IQ4_NL
- Ollama
How to use miceworks/Mark-SLM-V3-150M with Ollama:
ollama run hf.co/miceworks/Mark-SLM-V3-150M:IQ4_NL
- Unsloth Desktop
- Docker Model Runner
How to use miceworks/Mark-SLM-V3-150M with Docker Model Runner:
docker model run hf.co/miceworks/Mark-SLM-V3-150M:IQ4_NL
- Lemonade
How to use miceworks/Mark-SLM-V3-150M with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull miceworks/Mark-SLM-V3-150M:IQ4_NL
Run and chat with the model
lemonade run user.Mark-SLM-V3-150M-IQ4_NL
List all available models
lemonade list
- Atomic Chat
Sync MarkV3 HF AutoModel loader for 12h trained weights (configuration_mark_v3.py)
Browse files- configuration_mark_v3.py +4 -11
configuration_mark_v3.py
CHANGED
|
@@ -1,9 +1,4 @@
|
|
| 1 |
-
|
| 2 |
-
Hugging Face Transformers Configuration for Mark-SLM-V3-150M
|
| 3 |
-
"""
|
| 4 |
-
|
| 5 |
-
from transformers.configuration_utils import PretrainedConfig
|
| 6 |
-
|
| 7 |
|
| 8 |
class MarkV3Config(PretrainedConfig):
|
| 9 |
model_type = "deepseek2"
|
|
@@ -11,9 +6,9 @@ class MarkV3Config(PretrainedConfig):
|
|
| 11 |
|
| 12 |
def __init__(
|
| 13 |
self,
|
| 14 |
-
vocab_size: int =
|
| 15 |
hidden_size: int = 576,
|
| 16 |
-
num_hidden_layers: int =
|
| 17 |
num_attention_heads: int = 9,
|
| 18 |
num_key_value_heads: int = 9,
|
| 19 |
intermediate_size: int = 1536,
|
|
@@ -26,8 +21,7 @@ class MarkV3Config(PretrainedConfig):
|
|
| 26 |
num_experts_per_tok: int = 2,
|
| 27 |
n_shared_experts: int = 1,
|
| 28 |
moe_intermediate_size: int = 256,
|
| 29 |
-
max_position_embeddings: int =
|
| 30 |
-
original_max_position_embeddings: int = 4096,
|
| 31 |
rope_theta: float = 10000.0,
|
| 32 |
rope_scaling_factor: float = 2.0,
|
| 33 |
rms_norm_eps: float = 1e-6,
|
|
@@ -60,7 +54,6 @@ class MarkV3Config(PretrainedConfig):
|
|
| 60 |
self.n_shared_experts = n_shared_experts
|
| 61 |
self.moe_intermediate_size = moe_intermediate_size
|
| 62 |
self.max_position_embeddings = max_position_embeddings
|
| 63 |
-
self.original_max_position_embeddings = original_max_position_embeddings
|
| 64 |
self.rope_theta = rope_theta
|
| 65 |
self.rope_scaling_factor = rope_scaling_factor
|
| 66 |
self.rms_norm_eps = rms_norm_eps
|
|
|
|
| 1 |
+
from transformers import PretrainedConfig
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
|
| 3 |
class MarkV3Config(PretrainedConfig):
|
| 4 |
model_type = "deepseek2"
|
|
|
|
| 6 |
|
| 7 |
def __init__(
|
| 8 |
self,
|
| 9 |
+
vocab_size: int = 49152,
|
| 10 |
hidden_size: int = 576,
|
| 11 |
+
num_hidden_layers: int = 14,
|
| 12 |
num_attention_heads: int = 9,
|
| 13 |
num_key_value_heads: int = 9,
|
| 14 |
intermediate_size: int = 1536,
|
|
|
|
| 21 |
num_experts_per_tok: int = 2,
|
| 22 |
n_shared_experts: int = 1,
|
| 23 |
moe_intermediate_size: int = 256,
|
| 24 |
+
max_position_embeddings: int = 4096,
|
|
|
|
| 25 |
rope_theta: float = 10000.0,
|
| 26 |
rope_scaling_factor: float = 2.0,
|
| 27 |
rms_norm_eps: float = 1e-6,
|
|
|
|
| 54 |
self.n_shared_experts = n_shared_experts
|
| 55 |
self.moe_intermediate_size = moe_intermediate_size
|
| 56 |
self.max_position_embeddings = max_position_embeddings
|
|
|
|
| 57 |
self.rope_theta = rope_theta
|
| 58 |
self.rope_scaling_factor = rope_scaling_factor
|
| 59 |
self.rms_norm_eps = rms_norm_eps
|