GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
stancsz
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
stancsz
Model Tree
Input Modalities
Output Modalities
Supported Functionality
powershell
hf download stancsz/Wrench-4B-Qwen3.6-8E-NVFP4-native4M-v97-dense-native-gate-Experimental-Preview ` --local-dir Wrench-4B-Qwen3.6-8E-NVFP4-native4M-v97-dense-native-gateSet-Location Wrench-4B-Qwen3.6-8E-NVFP4-native4M-v97-dense-native-gate.\run_wrench.ps1The package is a Hugging Face Safetensors directory with tokenizer, bundled
verifier, deterministic toolbelt, context policy, and a model-local server.
It accepts Ollama-shaped /api/chat and /api/generate requests, including
options.num_ctx=4000000, without an external gateway.
This NVFP4 package is approximately 3.2 GiB on disk. The BF16 development package is much larger and is not the recommended download for normal users.
The production-value path is hybrid and model-local:
The first-layer receipt records selected and omitted spans, raw payload hash, effective working context, route source, and gate latency. Dense native 2M/4M attention is optional research. It is not the Wrench product claim.
When the optional dense-native lane is enabled, -NativeDirectInput enables
the first-layer gate automatically. -DenseNativeGate can also be supplied
explicitly. The raw request still arrives at the model-local native
endpoint, but the package's first model-side stage compacts it to 32K to 64K
before expensive attention:
powershell
.\serve_freetoken.ps1 -OllamaApi -NativeDirectInput -DenseNativeGateThe gate is fail-closed and records the raw payload hash, selected working
context, and gate stage. -BypassDenseNativeGate is reserved for separate
reducer-bypassed native capacity probes and cannot be combined with a
dense-native mode switch.
powershell
python .\wrench_server.py --model-dir . --allowed-root . --mechanical-onlyThe endpoint is http://127.0.0.1:28900/v1/chat/completions. The same process
also exposes /api/tags, /api/show, /api/chat, and /api/generate for
clients that expect an Ollama-shaped surface. It receives the full raw
conversation directly and emits hash-bound context-gate receipts.
The --mechanical-only mode is the verified fast path. Remove it only when a
compatible local native backend is available for ambiguous requests. Native
generation is separately verified and must not be inferred from the API shape.
The embedded worker API is also available directly from the downloaded directory:
python
from wrench_worker import WrenchWorker worker = WrenchWorker.from_pretrained( "./Wrench-4B-Qwen3.6-8E-NVFP4-native4M-v97-dense-native-gate", load_model=False)result = worker.propose([ {"role": "user", "content": "Read README.md with a 65536 byte limit."}])On the historical 220-case diagnostic replay, with client-side mechanical shortcut disabled:
94.5411%;95.5310%;99.6503%;183.314 ms / 337.174 ms;0;0.The current v97 NVFP4 package passed a direct model-local 4M probe with
3,999,995 estimated prompt tokens, HTTP 200, 172.15 ms elapsed, and zero
model calls. The broader historical matrix passed three repetitions each at
64K, 128K, 256K, 2M, and 4M. The 2M/4M reference-lookup probe recovered exact
proposals in 18/18 cases, with zero model calls.
These are hybrid model-local diagnostics, not dense native attention quality, stock Ollama native generation quality, family-disjoint approval, or production enablement. The current stock Ollama native generation boundary is explicitly recorded as failed on the validation host. GGUF and vLLM require architecture adapters and are not claimed as verified.
The full source regression is 176 passed. Final release still requires the
human-approved family-disjoint MiniMax-worker trace set, independent RTX 5060
Ti verification, and operational shadow evidence. Wrench has no direct
mutation authority. It proposes bounded actions or abstains, and the
surrounding verifier must enforce execution policy.
hf download stancsz/Wrench-4B-Qwen3.6-8E-NVFP4-native4M-v97-dense-native-gate-Experimental-Preview ` --local-dir Wrench-4B-Qwen3.6-8E-NVFP4-native4M-v97-dense-native-gateSet-Location Wrench-4B-Qwen3.6-8E-NVFP4-native4M-v97-dense-native-gate.\run_wrench.ps1from wrench_worker import WrenchWorker worker = WrenchWorker.from_pretrained( "./Wrench-4B-Qwen3.6-8E-NVFP4-native4M-v97-dense-native-gate", load_model=False)result = worker.propose([ {"role": "user", "content": "Read README.md with a 65536 byte limit."}])