Mamba
State-space model with O(n) complexity vs Transformers' O(n²). 5× faster inference, million-token sequences, no KV cache. Selective SSM with hardware-aware design. Mamba-1 (d_state=16) and Mamba-2 (d_state=128, multi-head). Models 130M-2.8B on HuggingFace.From its SKILL.md
npx -y skills add nota-america/forgecat-agent-profiles --skill mambaAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 24 stars24 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its file declares
Copied from the file, not written here
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
7.2 KB, ~1.9k tokens by cl100k_base, as published. Nobody here has run it
Mamba - Selective State Space Models
Quick start
Mamba is a state-space model architecture achieving O(n) linear complexity for sequence modeling.
Installation:
# Install causal-conv1d (optional, for efficiency)
pip install causal-conv1d>=1.4.0
# Install Mamba
pip install mamba-ssm
# Or both together
pip install mamba-ssm[causal-conv1d]
Prerequisites: Linux, NVIDIA GPU, PyTorch 1.12+, CUDA 11.6+
Basic usage (Mamba block):
import torch
from mamba_ssm import Mamba
batch, length, dim = 2, 64, 16
x = torch.randn(batch, length, dim).to("cuda")
model = Mamba(
d_model=dim, # Model dimension
d_state=16, # SSM state dimension
d_conv=4, # Conv1d kernel size
expand=2 # Expansion factor
).to("cuda")
y = model(x) # O(n) complexity!
assert y.shape == x.shape
Common workflows
Workflow 1: Language model with Mamba-2
Complete LM with generation:
from mamba_ssm.models.mixer_seq_simple import MambaLMHeadModel
from mamba_ssm.models.config_mamba import MambaConfig
import torch
# Configure Mamba-2 LM
config = MambaConfig(
d_model=1024, # Hidden dimension
n_layer=24, # Number of layers
vocab_size=50277, # Vocabulary size
ssm_cfg=dict(
layer="Mamba2", # Use Mamba-2
d_state=128, # Larger state for Mamba-2
headdim=64, # Head dimension
ngroups=1 # Number of groups
)
)
model = MambaLMHeadModel(config, device="cuda", dtype=torch.float16)
# Generate text
input_ids = torch.randint(0, 1000, (1, 20), device="cuda", dtype=torch.long)
output = model.generate(
input_ids=input_ids,
max_length=100,
temperature=0.7,
top_p=0.9
)
Workflow 2: Use pretrained Mamba models
Load from HuggingFace:
from transformers import AutoTokenizer
from mamba_ssm.models.mixer_seq_simple import MambaLMHeadModel
# Load pretrained model
model_name = "state-spaces/mamba-2.8b"
tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-neox-20b") # Use compatible tokenizer
model = MambaLMHeadModel.from_pretrained(model_name, device="cuda", dtype=torch.float16)
# Generate
prompt = "The future of AI is"
input_ids = tokenizer(prompt, return_tensors="pt").input_ids.to("cuda")
output_ids = model.generate(
input_ids=input_ids,
max_length=200,
temperature=0.7,
top_p=0.9,
repetition_penalty=1.2
)
generated_text = tokenizer.decode(output_ids[0])
print(generated_text)
Available models:
state-spaces/mamba-130mstate-spaces/mamba-370mstate-spaces/mamba-790mstate-spaces/mamba-1.4bstate-spaces/mamba-2.8b
Workflow 3: Mamba-1 vs Mamba-2
Mamba-1 (smaller state):
from mamba_ssm import Mamba
model = Mamba(
d_model=256,
d_state=16, # Smaller state dimension
d_conv=4,
expand=2
).to("cuda")
Mamba-2 (multi-head, larger state):
from mamba_ssm import Mamba2
model = Mamba2(
d_model=256,
d_state=128, # Larger state dimension
d_conv=4,
expand=2,
headdim=64, # Head dimension for multi-head
ngroups=1 # Parallel groups
).to("cuda")
Key differences:
- State size: Mamba-1 (d_state=16) vs Mamba-2 (d_state=128)
- Architecture: Mamba-2 has multi-head structure
- Normalization: Mamba-2 uses RMSNorm
- Distributed: Mamba-2 supports tensor parallelism
Workflow 4: Benchmark vs Transformers
Generation speed comparison:
# Benchmark Mamba
python benchmarks/benchmark_generation_mamba_simple.py \
--model-name "state-spaces/mamba-2.8b" \
--prompt "The future of machine learning is" \
--topp 0.9 --temperature 0.7 --repetition-penalty 1.2
# Benchmark Transformer
python benchmarks/benchmark_generation_mamba_simple.py \
--model-name "EleutherAI/pythia-2.8b" \
--prompt "The future of machine learning is" \
--topp 0.9 --temperature 0.7 --repetition-penalty 1.2
Expected results:
- Mamba: 5× faster inference
- Memory: No KV cache needed
- Scaling: Linear with sequence length
When to use vs alternatives
Use Mamba when:
- Need long sequences (100K+ tokens)
- Want faster inference than Transformers
- Memory-constrained (no KV cache)
- Building streaming applications
- Linear scaling important
Advantages:
- O(n) complexity: Linear vs quadratic
- 5× faster inference: No attention overhead
- No KV cache: Lower memory usage
- Million-token sequences: Hardware-efficient
- Streaming: Constant memory per token
Use alternatives instead:
- Transformers: Need best-in-class performance, have compute
- RWKV: Want RNN+Transformer hybrid
- RetNet: Need retention-based architecture
- Hyena: Want convolution-based approach
Common issues
Issue: CUDA out of memory
Reduce batch size or use gradient checkpointing:
model = MambaLMHeadModel(config, device="cuda", dtype=torch.float16)
model.gradient_checkpointing_enable() # Enable checkpointing
Issue: Slow installation
Install binary wheels (not source):
pip install mamba-ssm --no-build-isolation
Issue: Missing causal-conv1d
Install separately:
pip install causal-conv1d>=1.4.0
Issue: Model not loading from HuggingFace
Use MambaLMHeadModel.from_pretrained (not AutoModel):
from mamba_ssm.models.mixer_seq_simple import MambaLMHeadModel
model = MambaLMHeadModel.from_pretrained("state-spaces/mamba-2.8b")
Advanced topics
Selective SSM: See references/selective-ssm.md for mathematical formulation, state-space equations, and how selectivity enables O(n) complexity.
Mamba-2 architecture: See references/mamba2-details.md for multi-head structure, tensor parallelism, and distributed training setup.
Performance optimization: See references/performance.md for hardware-aware design, CUDA kernels, and memory efficiency techniques.
Hardware requirements
- GPU: NVIDIA with CUDA 11.6+
- VRAM:
- 130M model: 2GB
- 370M model: 4GB
- 790M model: 8GB
- 1.4B model: 14GB
- 2.8B model: 28GB (FP16)
- Inference: 5× faster than Transformers
- Memory: No KV cache (lower than Transformers)
Performance (vs Transformers):
- Speed: 5× faster inference
- Memory: 50% less (no KV cache)
- Scaling: Linear vs quadratic
Resources
- Paper (Mamba-1): https://arxiv.org/abs/2312.00752 (Dec 2023)
- Paper (Mamba-2): https://arxiv.org/abs/2405.21060 (May 2024)
- GitHub: https://github.com/state-spaces/mamba ⭐ 13,000+
- Models: https://huggingface.co/state-spaces
- Docs: Repository README and wiki
What ships with it: 3 files
22.1 KB alongside SKILL.md
references/
- architecture-details.md5.3 KB
- benchmarks.md7.9 KB
- training-guide.md8.8 KB
Gives 0 of the 12 instructions most architecture codebase skills give in ~1.9k tokens
Counted across 811 of the 1,134 authors here whose files we hold, read 2026-08-07
- Ask the user which candidate to explorein 45 of 811, across 15 files
- Apply the deletion test to suspected shallow modulesin 43 of 811, across 15 files
- Read any relevant architecture decision records firstin 31 of 811, across 8 files
- Use exact glossary terms in every suggestionin 30 of 811, across 10 files
- Accept dependencies instead of creating themin 24 of 811, across 5 files
- Include before and after visualisations for each candidatein 24 of 811, across 5 files
- Read the domain glossary before exploringin 24 of 811, across 6 files
- Return results instead of producing side effectsin 23 of 811, across 4 files
- Explore the codebase for shallow modules and frictionin 23 of 811, across 3 files
- Introduce seams only where things varyin 22 of 811, across 3 files
- Reduce the number of methodsin 21 of 811, across 2 files
- Design deep modules with small interfacesin 21 of 811, across 3 files
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.