FinanceGPT Developer Platform
Quantization, ONNX & runtime export
Prepare reproducible developer-side runtime exports from pinned transformer revisions. HF10 provides governed recipes and lineage manifests; it does not run conversion jobs or publish derived artifacts automatically.
Transformer models
1
Public Model Hub candidates
Optimization-capable
1
Pinned repository + revision
Derived exports published
0
HF10 never invents converted artifacts
Server-side jobs
0
Developer-side recipes only
Supported developer workflows
Optimization is treated as creation of a derived artifact with explicit source lineage and post-conversion evaluation.
ONNX export
Optimum + ONNX Runtime
Dynamic INT8
ONNX Runtime recipe
8-bit loading
bitsandbytes recipe
4-bit NF4
Optional hardware-sensitive recipe
Models
Only immutable source revisions are considered reproducible optimization bases.
| Model | Task | Revision | Exports | Quantization | Status | Action |
|---|---|---|---|---|---|---|
all-MiniLM-L6-v2 sentence-transformers/all-MiniLM-L6-v2 |
feature-extraction | ea78891063587eb050ed... |
onnx | onnx-dynamic-int8, bitsandbytes-int8, bitsandbytes-nf4 | Ready | Open recipes |