Gulshan Yadav
|
62ec3c92da
|
feat(sdk): add Flutter/Dart SDK for Synor Compute
Complete SDK implementation for Flutter and Dart applications:
lib/src/types.dart:
- Precision, ProcessorType, Priority, JobStatus enums
- SynorConfig for client configuration
- MatMulOptions, Conv2dOptions, AttentionOptions, InferenceOptions
- PricingInfo and UsageStats data classes
- SynorException for error handling
lib/src/tensor.dart:
- Full Tensor class with shape, dtype, and data
- Factory constructors: zeros, ones, rand, randn, eye, linspace, arange
- Operations: reshape, transpose, flatten
- Statistics: sum, mean, std, min, max, argmin, argmax
- Element-wise: add, sub, mul, div, scalar ops
- Activations: relu, sigmoid, tanh, softmax
- JSON serialization with base64-encoded binary data
lib/src/job.dart:
- JobResult with status, result, timing, and cost
- Job class with WebSocket streaming and HTTP polling
- JobStatusUpdate for real-time progress tracking
- JobBatch for parallel job management
lib/src/client.dart:
- SynorCompute main client
- Operations: matmul, conv2d, attention, elementwise, reduce
- LLM inference with streaming support
- Tensor upload/download/delete
- Job management: submit, cancel, list
- Pricing and usage statistics
Platform support: Android, iOS, Linux, macOS, Web, Windows
|
2026-01-11 14:27:55 +05:30 |
|
Gulshan Yadav
|
a808bb37a6
|
feat(sdk): add consumer SDKs for JavaScript, Python, and Go
Add complete SDK implementations for accessing Synor Compute:
JavaScript/TypeScript SDK (sdk/js/):
- Full async/await API with TypeScript types
- Tensor operations: matmul, conv2d, attention
- Model inference with streaming support
- WebSocket-based job monitoring
- Browser and Node.js compatible
Python SDK (sdk/python/):
- Async/await with aiohttp
- NumPy integration for tensors
- Context managers for cleanup
- Type hints throughout
- PyPI-ready package structure
Go SDK (sdk/go/):
- Idiomatic Go with context support
- Efficient binary tensor serialization
- HTTP client with configurable timeouts
- Zero external dependencies (stdlib only)
All SDKs support:
- Matrix multiplication (FP64 to INT4 precision)
- Convolution operations (2D, 3D)
- Flash attention
- LLM inference
- Spot pricing queries
- Job polling and cancellation
- Heterogeneous compute targeting (CPU/GPU/TPU/NPU/LPU)
|
2026-01-11 14:11:58 +05:30 |
|