blt/bytelatent/eval.py

260 lines
8.8 KiB
Python
Raw Normal View History

2024-12-12 23:32:30 +00:00
# Copyright (c) Meta Platforms, Inc. and affiliates.
import json
import logging
import os
from collections import defaultdict
from datetime import datetime
from pathlib import Path
from typing import Any
2024-12-12 23:32:30 +00:00
import torch
from lm_eval import simple_evaluate
from lm_eval.api.instance import Instance
from lm_eval.api.model import LM
from omegaconf import OmegaConf
from pydantic import BaseModel, ConfigDict
2024-12-12 23:32:30 +00:00
from bytelatent.args import EvalArgs, ValidationArgs, parse_args
2024-12-12 23:32:30 +00:00
from bytelatent.checkpoint import CONSOLIDATE_FOLDER, consolidate_checkpoints
from bytelatent.data.file_util import get_fs
2024-12-12 23:32:30 +00:00
from bytelatent.distributed import (
DistributedArgs,
dist_mean_dict,
get_global_rank,
get_world_size,
setup_torch_distributed,
)
from bytelatent.generate import (
2024-12-12 23:32:30 +00:00
PackedCausalTransformerGenerator,
load_consolidated_model_and_tokenizer,
)
from bytelatent.transformer import LMTransformer, LMTransformerArgs
2024-12-12 23:32:30 +00:00
EVAL_FOLDER_NAME = "{:010d}"
logger = logging.getLogger()
def all_dicts_same(dict_list):
if not dict_list: # Check if the list is empty
return True
# Compare each dictionary to the first one
first_dict = dict_list[0]
return all(d == first_dict for d in dict_list)
class MockAccelerator:
def gather(self, tensor):
l = [torch.zeros_like(tensor) for _ in range(get_world_size())]
torch.distributed.all_gather(l, tensor)
return torch.stack(l)
def wait_for_everyone(self):
torch.distributed.barrier()
# Light wrapper around generator for lm-eval harness
class EvalHarnessLM(LM):
def __init__(self, generator):
super().__init__()
self.generator = generator
self.accelerator = MockAccelerator()
self._rank = get_global_rank()
self._world_size = get_world_size()
self.device = generator.device
def generate_until(self, requests: list[Instance]) -> list[str]:
2024-12-12 23:32:30 +00:00
prompts, gen_args = zip(*[req.args for req in requests])
assert all_dicts_same(gen_args), "Doesn't support different gen args for now"
gen_args = gen_args[0]
temperature = gen_args.get("temperature", 0.0)
top_p = gen_args.get("top_p", None)
top_k = gen_args.get("top_k", None)
until = gen_args.get("until", [])
self.generator.temperature = temperature
self.generator.top_p = top_p
self.generator.top_k = top_k
self.generator.until = until
generations, _, _ = self.generator.generate(prompts)
filtered_gen = []
for g in generations:
for e in until:
g = g.replace(e, "")
filtered_gen.append(g)
return filtered_gen
def loglikelihood(self, requests: list[Instance]) -> list[tuple[float, bool]]:
2024-12-12 23:32:30 +00:00
prompts, continuations = zip(*[req.args for req in requests])
inputs = [req.args[0] + req.args[1] for req in requests]
max_gen_len = self.generator.max_gen_len
# We temporarily lower max gen len
self.generator.max_gen_len = 1
_, lls, greedy = self.generator.generate(inputs)
results = []
for p, ll, gr in zip(prompts, lls, greedy):
p_len = len(
self.generator.tokenizer.encode(p, add_bos=False, add_eos=False)
)
results.append((ll[p_len:].sum().item(), gr[p_len:].all().item()))
self.generator.max_gen_len = max_gen_len
return results
def loglikelihood_rolling(self, requests: list[Instance]) -> list[float]:
2024-12-12 23:32:30 +00:00
prompts = [req.args[0] for req in requests]
max_gen_len = self.generator.max_gen_len
# We temporarily lower max gen len
self.generator.max_gen_len = 1
_, lls, _ = self.generator.generate(prompts)
results = []
for ll in lls:
results.append((ll.sum().item(),))
self.generator.max_gen_len = max_gen_len
return results
def eval_on_val(generator, val_args: ValidationArgs, train_cfg):
srcs = {}
for src in val_args.sources:
path = os.path.join(val_args.root_dir, src)
srcs[path] = 1.0
for src in train_cfg.data.sources:
path = os.path.join(train_cfg.data.root_dir, src)
srcs[path] = 1.0
multi_state = init_choice_state(
"", srcs, 0, get_global_rank(), get_world_size(), "*.val.jsonl"
)
path_to_iter = setup_sources(multi_state)
max_gen_len = generator.max_gen_len
# We temporarily lower max gen len
generator.max_gen_len = 1
all_val_metrics = {}
for src in path_to_iter:
jsonl_iterator = path_to_iter[src]
texts = []
logger.info(f"Running validation on {src}...")
for step, (content, state) in enumerate(jsonl_iterator):
if state["current_iter"] > 0 or (
val_args.max_steps is not None and step >= val_args.max_steps
):
break
content_key = "text" if ("text" in content) else "content"
texts.append(content[content_key])
_, loglikelihood, _ = generator.generate(texts)
metrics = defaultdict(list)
for i, ll in enumerate(loglikelihood):
tmp = ll.sum().item()
metrics["nll"].append(tmp)
metrics["nll_per_token"].append(tmp / len(ll))
metrics["nll_per_char"].append(tmp / len(texts[i]))
metrics["avg_seqlen"].append(len(ll))
for m in metrics:
metrics[m] = sum(metrics[m]) / len(metrics[m])
metrics.update(dist_mean_dict(metrics))
logger.info(f"Validation on {src} done. Metrics: {metrics}")
name = os.path.basename(src)
if name in all_val_metrics:
logger.warning(
f"Duplicate source name {name}, path {src} in validation sources, renaming to {name}_1"
)
name = f"{name}_1"
all_val_metrics[name] = metrics
generator.max_gen_len = max_gen_len
return all_val_metrics
def launch_eval(eval_args: EvalArgs):
2024-12-12 23:32:30 +00:00
if not torch.distributed.is_initialized():
setup_torch_distributed(DistributedArgs())
fs = get_fs(eval_args.ckpt_dir, s3_profile=eval_args.s3_profile)
2024-12-12 23:32:30 +00:00
if (
fs.exists(eval_args.ckpt_dir)
and fs.exists(os.path.join(eval_args.ckpt_dir, "params.json"))
and len(fs.glob(os.path.join(eval_args.ckpt_dir, "*.pth"))) != 0
2024-12-12 23:32:30 +00:00
):
consolidate_path = eval_args.ckpt_dir
2024-12-12 23:32:30 +00:00
else:
consolidate_path = os.path.join(eval_args.ckpt_dir, CONSOLIDATE_FOLDER)
if not fs.exists(consolidate_path) and get_global_rank() == 0:
consolidate_path = consolidate_checkpoints(eval_args.ckpt_dir)
2024-12-12 23:32:30 +00:00
fs.mkdirs(eval_args.dump_dir, exist_ok=True)
with fs.open(os.path.join(eval_args.dump_dir, "config.yaml"), "w") as f:
f.write(eval_args.model_dump_json())
2024-12-12 23:32:30 +00:00
torch.distributed.barrier()
logger.info("Loading model")
# TODO: Make this general so that it works with either
# LMTransformer or Blt, similar with args
2024-12-12 23:32:30 +00:00
model, tokenizer, train_cfg = load_consolidated_model_and_tokenizer(
consolidate_path,
)
logger.info("Model loaded")
model.eval()
generator = PackedCausalTransformerGenerator(eval_args.generator, model, tokenizer)
2024-12-12 23:32:30 +00:00
wrap = EvalHarnessLM(generator)
# Redo
results = simple_evaluate(wrap, eval_args.harness.model_dump())
2024-12-12 23:32:30 +00:00
val_results = None
if eval_args.validation:
val_results = eval_on_val(generator, eval_args.validation, train_cfg)
2024-12-12 23:32:30 +00:00
if get_global_rank() == 0:
with fs.open(os.path.join(eval_args.dump_dir, "results.json"), "w") as f:
2024-12-12 23:32:30 +00:00
f.write(json.dumps(results))
logger.info(f"All evaluation results: {results['results']}")
if val_results is not None:
with fs.open(os.path.join(eval_args.dump_dir, "validation.json"), "w") as f:
2024-12-12 23:32:30 +00:00
f.write(json.dumps(val_results))
logger.info(f"All validation results: {val_results}")
if eval_args.metric_log_dir and get_global_rank() == 0:
metric_log_path = os.path.join(eval_args.metric_log_dir, "metrics.eval.jsonl")
2024-12-12 23:32:30 +00:00
logger.info(f"Writing metric logs to {metric_log_path}")
timestamp = {
"created_at": datetime.utcnow().isoformat(),
}
if eval_args.global_step is not None:
timestamp["global_step"] = eval_args.global_step
2024-12-12 23:32:30 +00:00
print(
json.dumps(timestamp | results["results"]),
file=fs.open(metric_log_path, mode="a"),
2024-12-12 23:32:30 +00:00
flush=True,
)
val_log_path = os.path.join(
eval_args.metric_log_dir, "metrics.validation.jsonl"
)
2024-12-12 23:32:30 +00:00
if val_results is not None:
print(
json.dumps(timestamp | val_results),
file=fs.open(val_log_path, mode="a"),
2024-12-12 23:32:30 +00:00
flush=True,
)
del generator
def main():
eval_args = parse_args(EvalArgs)
launch_eval(eval_args)
2024-12-12 23:32:30 +00:00
if __name__ == "__main__":
main()