| { | |
| "09-14-22": [ | |
| { | |
| "model_id": "bert-base-uncased", | |
| "download": 31330425 | |
| }, | |
| { | |
| "model_id": "Jean-Baptiste/camembert-ner", | |
| "download": 20521361 | |
| }, |
| { | |
| "roberta-base": { | |
| "model_id": "roberta-base", | |
| "size": 0.124645632, | |
| "code": 0, | |
| "memo": null | |
| }, | |
| "tals/albert-xlarge-vitaminc-mnli": { | |
| "model_id": "tals/albert-xlarge-vitaminc-mnli", | |
| "size": 0.058724864, |
This table documents the top-5000 most download HuggingFace models (during 8/8/2022-9/8/2022) sorted by their sizes. Note that some models that are not Huggingface transformers compatible won't be listed here.
| Model | #Parameters |
|---|---|
| bigscience/bloom | 176.2B |
| bigscience/bloom-petals | 176.2B |
| facebook/opt-66b | 65.7B |
| #!/usr/bin/env bash | |
| # The entry point of AWS Batch job. This script is in charge of configuring | |
| # the repo, executing the given command, and uploading the results. | |
| set -e | |
| date | |
| # Parse arguments | |
| SOURCE_REF=$1 | |
| REPO=$2 |
| #!/usr/bin/env bash | |
| set -e | |
| RUNNER_VERSION="2.287.1" | |
| # The path to the new runner. | |
| RUNNER_PATH=$1 | |
| # The target Github org. | |
| GITHUB_ORG=$2 | |
| # Optional runner label. |
| import numpy as np | |
| import tvm | |
| from tvm import relay, auto_scheduler | |
| import tvm.relay.testing | |
| from tvm.contrib import graph_executor | |
| def get_network(name, batch_size, layout="NHWC", dtype="float32"): | |
| """Get the symbol definition and random weight of a network""" |
| import numpy as np | |
| import tvm | |
| from tvm import te, tir, topi | |
| from tvm.topi import utils | |
| dev = tvm.device("gpu", 0) | |
| target = tvm.target.Target("cuda") | |
| ### Copy from topi/cuda/injective.py and make block/thread num configurable |
| home/ubuntu/meta/src/op/dispatch/tvmjit/unary.cc:55: Error: Failed to JIT mnm_op_erf: RuntimeError: | |
| #if defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 530) | |
| #include <cuda_fp16.h> | |
| __device__ half max(half a, half b) | |
| { | |
| return __hgt(__half(a), __half(b)) ? a : b; | |
| } | |
| __device__ half min(half a, half b) | |
| { | |
| return __hlt(__half(a), __half(b)) ? a : b; |
| # Pytorch reference | |
| # https://pytorch.org/tutorials/beginner/blitz/cifar10_tutorial.html#train-the-network | |
| import numpy as np | |
| import torch | |
| import torchvision | |
| import torchvision.transforms as transforms | |
| import tvm |
| import timeit | |
| import numpy as np | |
| import torch | |
| import tvm | |
| from tvm import auto_scheduler | |
| import mnm | |
| from mnm.testing.utils import ir_fusion, ir_simplify, get_vm_executor, get_vm_profiler |