jsgrad Docs Chat

Docs

Cloud Getting Started

API

Index type TensorOptions type TensorIndice type Layer type LayerAsync class Tensor function id variable sorted class DefaultMap class ArrayMap class WeakKeyMap class WeakValueMap class NotImplemented variable floatString variable constToNumeric variable max variable min variable abs variable trunc variable sqrt variable sin variable bigint_sqrt variable bigint_sin variable next variable int_to_bytes variable bytes_to_bigint variable isInf class Enum variable random_id variable string_to_bytes variable bytes_to_string variable bytes_to_hex variable concat_bytes function product variable divmod function counter variable list_str variable entries function is_less_than type ConstType variable isConst variable is_eq variable intersection function set_default function zip function range variable tuple variable assert function permutations function is_subset function math_gcd variable dedup variable argsort variable all_same variable isInt variable all_int variable colored variable colorize_float variable memsize_to_str variable ansistrip variable ansilen variable make_tuple variable flatten variable fully_flatten variable strip_parens variable round_up variable lo32 variable hi32 variable data64 variable data64Le variable getbits variable i2u variable merge_dicts function merge_maps function merge_sets variable partition variable get_single_element variable unwrap variable getChild variable word_wrap variable to_function_name class Metadata variable _METADATA class GlobalCounters variable perf variable round variable Timing variable _format_fcn class Profiling variable _ensure_downloads_dir variable cpu_time_execution variable cpu_objdump variable capstone_flatdump function replace type Slice function slice function cache function cache_fn type Math variable num variable add variable sub variable mul variable div variable idiv variable neg variable mod variable and variable or variable xor variable lshift variable rshift variable lt variable gt variable le variable ge variable ne variable eq variable polyN variable prod variable sum variable ceildiv variable pow function pairwise function accumulate variable vars type FmtStr variable bitcast class DType class PtrDType class ImageDType class dtypes type DTypeLike variable to_dtype variable promoLattice variable _get_recursive_parents variable least_upper_dtype variable least_upper_float variable DTYPES_DICT variable INVERSE_DTYPES_DICT variable sum_acc_dtype variable truncate type Variable type ConstLike class MathTrait class Ops class GroupOp variable view_supported_devices variable identity_element variable can_pad variable END_FOR_UOP variable resolve variable smax variable smin variable ssimplify variable sym_infer class UOp class KernelInfo variable python_alu variable exec_alu variable print_uops type UPatInput type UPatFn type Pattern class UPat class UPatAny class PatternMatcher class TrackedGraphRewrite class TrackedPatternMatcher variable launch_viz class RewriteContext variable graph_rewrite variable graph_rewrite_map variable spec variable type_verify function split_uop variable div_and_mod_folding variable canonicalize_simplex variable is_increasing variable parse_valid variable uop_given_valid variable simplify_valid variable sint_to_uop variable symbolic_simple variable symbolic variable symbolic_flat variable _substitute variable renderer type sint variable merge_views variable view_left variable TRANSCENDENTAL_SUPPORTED_DTYPES variable _lazy_map_numbers variable mantissa_bits variable exponent_bias variable exponent_mask variable shr variable shl variable rintk variable pow2if variable ilogb2k variable ldexp3k variable ldexp2k variable frexp variable payne_hanek_reduction variable cody_waite_reduction variable trig_poly variable sin_poly variable sin_poly_small variable sin_poly_large variable xsin variable xexp2 variable xlog2 variable fold_expanded variable fix_unfoldable_image_load variable buf_idx_pat variable float4_folding variable simplify_valid_load variable get_late_rewrite_patterns variable threefry2x32 variable sigmoid_like variable loop_collapse variable index_collapse variable gep_through_wmma variable no_vectorized_wmma variable reduce_collapse variable acc_pat variable rng_pat variable rng_aug variable index_load variable arange_augrng variable arange_m variable mulacc_unrolled variable sym variable _expand_arg_to_idx variable _choices_from_args variable _swizzle_args variable do_expand variable do_contract variable no_vectorized_alu variable create_gate variable expander variable no_vectorized_load_store variable no_vectorized_acc variable devectorize variable delete_redundant_gates variable load_store_indexing variable migrate_indexing variable move_mask variable pm_render variable full_graph_rewrite class MNIST variable precompute_freqs_cis variable complex_mult variable apply_rotary_emb variable repeat_kv class Attention class FeedForward class TransformerBlock variable sample class Transformer variable convert_from_huggingface variable convert_from_gguf variable fix_bf16 type Llama3Size type Llama3Quantize type Llama3Constructor type Llama3Load type Llama3StaticLoad type Llama3Message type Llama3Chat type Llama3StopReason type Llama3Usage type Llama3Response class Llama3 variable LANGUAGES variable MODELS type WhisperModel class MultiHeadAttention class ResidualAttentionBlock class AudioEncoder variable prep_audio variable init_whisper variable load_file_waveform variable transcribe_file class Tokenizer class _Device variable Device variable uop_buffer variable uop_realized variable uop_is_realized class Buffer variable is_dtype_supported class BufferSpec class Allocator class LRUAllocator class _MallocAllocator variable MallocAllocator variable MAP_JIT type ProgramCallArgs class Program class CPUProgram class CompileError class Compiler class ProfileEvent class ProfileDeviceEvent class ProfileRangeEvent class ProfileGraphEntry class ProfileGraphEvent class ProfileResult class Compiled class Model class BatchNorm class BatchNorm2d class BatchNorm3d class Conv2d class Conv1d class ConvTranspose2d class ConvTranspose1d class Linear class GroupNorm class InstanceNorm class LayerNorm class LayerNorm2d class RMSNorm class Embedding class LSTMCell class Optimizer class OptimizerGroup variable SGD class LARS variable AdamW variable Adam variable inverse_safe_dtypes variable safe_load_metadata variable safe_load variable safe_save variable get_state_dict variable get_parameters variable replace_state_dict variable load_state_dict variable tar_extract variable ggml_data_to_tensor variable gguf_load variable mnist variable cifar class GraphException variable apply_graph_to_jit variable get_input_replace class GraphRunner class MultiGraphRunner variable update_depends class CapturedJit variable _prepare_jit_inputs class TinyJit type RenderBarOptions type TqdmProgress type TqdmOnProgress type TqdmOptions class Tqdm variable env variable setEnv class MemoryView

Model

Abstract model to simplify calling, loading and saving different models. You only need to implement list of layers to create a new model

import { Layer, Tensor } from "@jsgrad/jsgrad"
export class MNIST extends Model {
 layers: Layer[] = [
   new Conv2d(1, 32, 5),
   Tensor.relu,
   // ...other layers
 ]
}
const mnist = new MNIST()

// call the model with some input
mnist.call(Tensor.rand([1, 1, 28, 28]))

// load weigths from ./mnist.safetensors
await mnist.load("./mnist.safetensors")

//save weigths to ./mnist.safetensors
await mnist.save("./mnist.safetensors")

abstract class Model {

DEFAULT_LOAD = string

abstract layers = Layer[]

Call model with a Tensor input, returns a Tensor with output.

import { MNIST, Tensor } from "@jsgrad/jsgrad"

const model = new MNIST()
const res = model.call(Tensor.rand([1, 1, 28, 28]))
console.log(await res.tolist())

call = (x: Tensor) => unknown

Load model weights from a .safetensors file at the given path or absolute URL

import { MNIST } from "@jsgrad/jsgrad"
const model = new MNIST()
await model.load("./model.safetensors")

load = (path?: string | Tensor, onProgress?: TqdmOnProgress) => unknown

Save model weights to a .safetensors file at the given path

import { MNIST } from "@jsgrad/jsgrad"

const model = new MNIST()
await model.save("./model.safetensors")

save = (path: string) => unknown

}