andreaskoepf

Created	aliases	References
2025-01-10	Towards System 2 Reasoning in LLMs	https://arxiv.org/abs/2501.04682

Towards System 2 Reasoning in LLMs: Learning How to Think With Meta Chain-of-Thought

andreaskoepf

"""
Modified version of https://github.com/Dao-AILab/flash-attention/blob/87a1277653fc55cd615f5341255e00c69d5c00a1/flash_attn/flash_attn_triton.py
Experiments with attention bias by andreas.koepf

Main fix was "fixing the fix", e.g. removing lines of the original like:
```
# BUG: have to store and immediately load
# tl.store(t_ptrs, o_scale)
# o_scale = tl.load(t_ptrs)
```

andreaskoepf

"""
*Experimental* implementation of FlashAttention in Triton.
Tested with triton==2.0.0.dev20221202.
Triton 2.0 has a new backend (MLIR) but seems like it doesn't yet work for head dimensions
other than 64:
https://github.com/openai/triton/blob/d376020f90002757eea3ea9475d4f7cfc2ec5ead/python/triton/ops/flash_attention.py#L207
We'll update this implementation with the new Triton backend once this is fixed.

We use the FlashAttention implementation from Phil Tillet a starting point.
https://github.com/openai/triton/blob/master/python/tutorials/06-fused-attention.py

andreaskoepf

from datasets import load_dataset

ds = load_dataset("/path/oasst1", name='ready')
train = ds['train']
val = ds['validation']
print(f'{len(train)=}')
print(f'{len(val)=}')

for i in range(5):
   print(train[i]["message_tree_id"])

andreaskoepf

import argparse
import sys
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer


def parse_args():
    parser = argparse.ArgumentParser()
    parser.add_argument("model_name", type=str, help="checkpoint path or model name")
    parser.add_argument("--dtype", type=str, default="auto", help="auto, fp16, bf16 or fp32")

andreaskoepf

You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct.
If you don't know the answer to a question, please don't share false information."

andreaskoepf

import argparse
import sys

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer


def parse_args() -> argparse.Namespace:
    parser = argparse.ArgumentParser(
        description="Push checkpoints in HF transformers format to the Huggingface Hub.",

andreaskoepf

from typing import Optional
import torch


def precompute_freqs_cis(
    dim: int, end: int, theta: float = 10000.0, scaling_factor: float = 1.0
) -> torch.Tensor:
    freqs = 1.0 / (theta ** (torch.arange(0, dim, 2).float() / dim))
    t = torch.arange(end, device=freqs.device).float() / scaling_factor  # type: ignore
    freqs = torch.outer(t, freqs).float()  # type: ignore

andreaskoepf

# sent to me by tju01, thx

# install base tools
apt update
apt install protobuf-compiler libssl-dev gcc pkg-config g++ make 

# install rust
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
source "$HOME/.cargo/env"

andreaskoepf

# adapted from: https://github.com/lucidrains/vit-pytorch/blob/main/vit_pytorch/vit.py

import torch
import torch.nn.functional as F
from torch import nn
from torch.utils import checkpoint

from einops import rearrange, repeat

import triton