Instructions to use WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B-exp with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B-exp with Transformers:

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

pipe = pipeline("text-generation", model="WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B-exp")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B-exp")
model = AutoModelForCausalLM.from_pretrained("WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B-exp", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B-exp with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B-exp"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B-exp",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B-exp

SGLang

How to use WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B-exp 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 "WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B-exp" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B-exp",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B-exp" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B-exp",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B-exp with Docker Model Runner:
```
docker model run hf.co/WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B-exp
```

Experimental — likely broken (renamed to -exp on 2026-08-12). This variant was produced by the cascaded REAM line (REAM160 → REAM128). Cascaded variants later showed severe multi-turn degradation: generation typically collapses from the second request onward. No quality evidence was ever collected for this release. Kept for archival reference only; not recommended for use. A rebuilt line (REAM144 / REAM96, built in a single step from the base model) is in progress.

DeepSeek V4 Flash 0731 REAM128-146B

This release is a calibration-guided, training-free cascade from the committed REAM160 checkpoint. It reduces every routed MoE layer from 160 to 128 experts while preserving 43 transformer layers, one shared expert, and top-k 6 routing. The nominal logical parameter count is 145,796,990,551.

Lineage and precision

Source: WaveCut/DeepSeek-V4-Flash-0731-REAM160-180B@7b7e03914a2cf1d676161ae18bf0f24c54adb931.
Run: 20260810T115212Z.
Build code revision: ec9da81a9bb12979973d834ffbe7dbf0a6292bad.
Routed expert weights retain the source-native E2M1 FP4 layout with E8M0 scales.
Attention projections and unchanged tensors retain the source FP8/mixed layout.
REAM accumulation is deterministic FP32; BF16 is transient compute, not the primary checkpoint format.
MTP is disabled (num_nextn_predict_layers: 0).
Checkpoint: 82,392,019,049 bytes across 44 Safetensors shards, config.json, and model.safetensors.index.json.
Committed NATIVE_READY.json SHA-256: 8c32c76316574c75f18b45be69bd3b54c0c547531249b02f424a66485dc4820e.

REAM methodology

REAM128 does not apply gradient updates or add training examples to the checkpoint. Fresh layerwise saliency statistics, expert groups, merges, router mappings, and activations were computed against the immutable REAM160 source.

Calibration composition: 3,072 samples x 512 tokens.

Slice	Samples	Share	Pinned sources and focus
Code	614	20.0%	`sahil2801/CodeAlpaca-20k` and `bigcode/self-oss-instruct-sc2-exec-filter-50k`; instruction following and executable code
Math	307	10.0%	`openai/gsm8k` and `AI-MO/NuminaMath-CoT`; arithmetic and mathematical reasoning
Agent/tool	1,075	35.0%	`NousResearch/hermes-function-calling-v1`; function selection and structured tool calls
Russian/Cyrillic	1,076	35.0%	`ZeroAgency/ru-big-russian-dataset`; Russian and Cyrillic language coverage

At each layer, REAP saliency is the routed activation norm multiplied by router probability. The 128 highest-saliency source experts become centroids; the remaining 32 source experts are assigned using output and router-gate similarity. Each group is merged by saliency-weighted deterministic FP32 accumulation, router rows are remapped, and top-k 6 routing is preserved.

Validation and integrity

Build hardware: 4x NVIDIA GeForce RTX 4090 24 GB.
Structural gate: 43 layers, 128 routed experts, one shared expert, and top-k 6.
Native release gate: all 46 checkpoint artifacts match the committed byte sizes and SHA-256 values.
Comparative heldout PPL/NLL: not measured for this cascade.
Semantic generation suite: not run for this cascade.
Runtime benchmark: not run for this cascade.

The integrity checks prove lineage and artifact identity; they are not a claim of semantic equivalence to REAM160.

Loading and runtime compatibility

The configuration and tokenizer can be inspected with Transformers:

from transformers import AutoConfig, AutoTokenizer

repo_id = "WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B"
config = AutoConfig.from_pretrained(repo_id, trust_remote_code=False)
tokenizer = AutoTokenizer.from_pretrained(repo_id, trust_remote_code=False)
print(config.num_hidden_layers, config.n_routed_experts, config.num_experts_per_tok)

Stock AutoModelForCausalLM, vLLM, and SGLang generation is not claimed for this packed E2M1/E8M0 checkpoint. A runtime must implement the 128-expert DeepSeek V4 topology and its native FP4/FP8 storage.

Limitations

Quality and throughput must not be inferred from the REAM160 release; this cascade has no PPL, NLL, semantic, or benchmark result.
This is a specialized mixed-precision checkpoint, not a conventional BF16/F16 Transformers checkpoint.
Hardware memory requirements depend on the runtime, sharding strategy, context length, and KV-cache configuration.

License

MIT License, following the source checkpoint.

Downloads last month: 118

Safetensors

Model size

146B params

Tensor type

BF16

F32

F8_E4M3

I64

Model tree for WaveCut/DeepSeek-V4-Flash-0731-REAM128-146B-exp

Base model

deepseek-ai/DeepSeek-V4-Flash-0731

Quantized

WaveCut/DeepSeek-V4-Flash-0731-REAM160-180B

Quantized

(2)

this model

Quantizations

2 models