vwxyzjn

import argparse
import random
import time
import os
import torch
import torch.distributed as dist
import numpy as np
from functools import partial
from typing import Optional

vwxyzjn

import argparse
import os
import time
import torch
import torch.distributed as dist

# noinspection PyUnresolvedReferences
import deep_ep
from utils import init_dist, bench, bench_kineto, calc_diff, create_grouped_scores, inplace_unique, per_token_cast_to_fp8, per_token_cast_back, hash_tensor

vwxyzjn

import os
import torch
import torch.distributed as dist
from typing import Callable, List, Tuple, Optional, Union

# noinspection PyUnresolvedReferences
import deep_ep_cpp
# noinspection PyUnresolvedReferences
from deep_ep_cpp import Config, EventHandle
from .utils import EventOverlap, check_nvlink_connections

vwxyzjn

import time
from vllm import LLM, SamplingParams
from vllm.outputs import PoolingRequestOutput, RequestOutput
from typing import Union
import threading
from threading import Event

class MyLLM(LLM):
    def keep_running(
        self,

vwxyzjn

import torch

# set seed
torch.manual_seed(42)

B, T, D = 1, 4, 10 # batch size, sequence length, vocab size
tensor = torch.rand(B, T, D, requires_grad=True)
labels = torch.Tensor([[1, 2, 3, 4]]).long()

print("="*100)

vwxyzjn

"""
# Convert HF to TorchTitan DCP
python convert.py hf_to_dcp --input-path meta-llama/Meta-Llama-3.1-8B --output-path ./torchtitan_model

# Convert TorchTitan DCP to HF (works with any checkpoint structure)
python convert.py dcp_to_hf --input-path ./torchtitan_model --output-path ./hf_model

# Model structure
If you run the following code, you will get the model structure.

vwxyzjn

absl-py==2.3.0
aiohappyeyeballs==2.6.1
aiohttp==3.12.13
aiosignal==1.3.2
annotated-types==0.7.0
anyio==4.9.0
attrs==25.3.0
azure-core==1.34.0
azure-identity==1.23.0
azure-storage-blob==12.25.1

vwxyzjn

# SPDX-License-Identifier: Apache-2.0
"""Example for starting a Gradio OpenAI Chatbot Webserver
Start vLLM API server:
    vllm serve allenai/OLMo-2-0425-1B-Instruct

Start Gradio OpenAI Chatbot Webserver:
    python x1.py -m allenai/OLMo-2-0425-1B-Instruct --model-url http://ceres-cs-aus-441:8000/v1

Note that `pip install --upgrade gradio` is needed to run this example.
More details: https://github.com/gradio-app/gradio

vwxyzjn

for seed in 1 2; do
for lr in 5e-7 7e-7 9e-7; do
python update_command_args.py scripts/train/olmo2/grpo_7b.sh \
    --priority urgent \
    --workspace ai2/olmo-instruct \
    --exp_name 0423_grpo_seed_${seed}_lr_${lr} \
    --model_name_or_path allenai/OLMo-2-0425-1B-DPO \
    --model_revision main \
    --tokenizer_name_or_path allenai/OLMo-2-1124-7B-DPO \

vwxyzjn

TEMPLATE = """
---
license: apache-2.0
language:
- en
datasets:
- {{dataset}}
base_model:
- {{base_model}}
pipeline_tag: text-generation