grok-1/runners.py

310 lines
10 KiB
Python
Raw Normal View History

2024-03-15 01:03:58 +03:00
# Copyright 2024 X.AI Corp.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import bisect
import functools
import logging
import math
2024-08-08 02:54:48 +03:00
import numpy as np
2024-03-15 01:03:58 +03:00
import jax
import jax.numpy as jnp
2024-08-08 02:54:48 +03:00
import haiku as hk
2024-03-15 01:03:58 +03:00
import sentencepiece
2024-08-08 02:54:48 +03:00
from dataclasses import dataclass
from typing import Any, Callable, NamedTuple, Optional, Tuple
from jax.experimental import pjit
2024-03-15 01:03:58 +03:00
from jax.sharding import PartitionSpec as P
from model import (
LanguageModelConfig,
LanguageModelOutput,
TrainingState,
apply_rules,
Memory,
KVMemory,
)
2024-08-08 02:54:48 +03:00
import checkpoint as xai_checkpoint
2024-03-15 01:03:58 +03:00
logger = logging.getLogger(__name__)
2024-08-08 02:54:48 +03:00
logger.setLevel(logging.INFO)
2024-03-15 01:03:58 +03:00
rank_logger = logging.getLogger("rank")
2024-08-08 02:54:48 +03:00
rank_logger.setLevel(logging.INFO)
2024-03-15 01:03:58 +03:00
TOP_K = 8
class SampleSettings(NamedTuple):
2024-08-08 02:54:48 +03:00
temperature: jax.Array
nucleus_p: jax.Array
mask: jax.Array
active: jax.Array
2024-03-15 01:03:58 +03:00
class SampleOutput(NamedTuple):
2024-08-08 02:54:48 +03:00
token_id: jax.Array
prob: jax.Array
top_k_token_ids: jax.Array
top_k_probs: jax.Array
2024-03-15 01:03:58 +03:00
2024-08-08 02:54:48 +03:00
def insert_slice(memory: Memory, slice: Memory, length: int, i: int) -> Memory:
2024-03-15 01:03:58 +03:00
slice = Memory(
layers=[
KVMemory(layer.k, layer.v, step=jnp.array([length]))
for layer in slice.layers
],
)
return jax.tree_map(lambda m, u: jax.lax.dynamic_update_index_in_dim(m, u[0], i, axis=0),
memory, slice)
2024-08-08 02:54:48 +03:00
def pad_to_size(x: jnp.ndarray, size: int) -> jnp.ndarray:
2024-03-15 01:03:58 +03:00
if x.shape[0] > size:
x = x[-size:]
return np.pad(x, [0, size - x.shape[0]], mode="constant", constant_values=0)
def top_p_filter(logits: jax.Array, top_p: jax.Array) -> jax.Array:
assert logits.ndim == top_p.ndim, f"Expected {logits.ndim} equal {top_p.ndim}"
sorted_logits = jax.lax.sort(logits, is_stable=False)
sorted_probs = jax.nn.softmax(sorted_logits)
threshold_idx = jnp.argmax(jnp.cumsum(sorted_probs, -1) >= 1 - top_p, axis=-1)
threshold_largest_logits = jnp.take_along_axis(
sorted_logits, threshold_idx[..., jnp.newaxis], axis=-1
)
assert threshold_largest_logits.shape == logits.shape[:-1] + (1,)
mask = logits >= threshold_largest_logits
logits = jnp.where(mask, logits, -1e10)
return logits
def sample_token(
rngs: jax.random.PRNGKey,
lm_outputs: LanguageModelOutput,
settings: SampleSettings,
) -> SampleOutput:
settings = SampleSettings(
2024-08-08 02:54:48 +03:00
temperature=jnp.expand_dims(settings.temperature, (1, 2)),
nucleus_p=jnp.expand_dims(settings.nucleus_p, (1, 2)),
mask=jnp.expand_dims(settings.mask, 1),
active=settings.active,
2024-03-15 01:03:58 +03:00
)
logits = lm_outputs.logits / settings.temperature.astype(lm_outputs.logits.dtype)
logits = jnp.where(settings.mask, logits, -1e10)
logits = top_p_filter(logits, settings.nucleus_p.astype(logits.dtype))
new_token = jax.vmap(jax.random.categorical)(rngs, logits)
probabilities = jax.nn.softmax(logits)
token_prob = jnp.take_along_axis(probabilities, jnp.expand_dims(new_token, 1), axis=2)
token_prob = jnp.squeeze(token_prob, 1)
top_k_probs, top_k_token_ids = jax.lax.top_k(probabilities, TOP_K)
top_k_probs = jnp.squeeze(top_k_probs, 1)
top_k_token_ids = jnp.squeeze(top_k_token_ids, 1)
return SampleOutput(
new_token,
token_prob,
top_k_token_ids,
top_k_probs,
)
@dataclass
class ModelRunner:
model: LanguageModelConfig
bs_per_device: float = 2.0
load_rename_rules: Optional[list[tuple[str, str]]] = None
load_exclude_rules: Optional[list[str]] = None
2024-08-08 02:54:48 +03:00
rng_seed: int = 42
2024-03-15 01:03:58 +03:00
transform_forward: bool = False
checkpoint_path: str = ""
def make_forward_fn(self, mesh: Any):
def forward(tokens):
out = self.model.make(mesh=mesh)(tokens)
return out, None
if self.transform_forward:
forward = hk.transform(forward)
return forward
def initialize(
self,
init_data,
2024-08-08 02:54:48 +03:00
local_mesh_config: Tuple[int, int],
between_hosts_config: Tuple[int, int],
2024-03-15 01:03:58 +03:00
):
num_replicas = math.prod(between_hosts_config)
self.model.initialize()
self.model.fprop_dtype = jnp.bfloat16
num_local_gpus = len(jax.local_devices())
self.batch_size = int(self.bs_per_device * num_local_gpus * num_replicas)
self.local_batch_size = self.batch_size // jax.process_count()
self.local_mesh_config = local_mesh_config
self.between_hosts_config = between_hosts_config
rank_logger.info(
f"Initializing mesh for {self.local_mesh_config=} {self.between_hosts_config=}..."
)
self.mesh = make_mesh(self.local_mesh_config, self.between_hosts_config)
self.forward = self.make_forward_fn(mesh=self.mesh)
self.logits_fn = hk.transform(lambda tokens: self.forward(tokens)[0])
self.eval_forward = self.make_forward_fn(mesh=self.mesh)
self.logits_eval_fn = hk.transform(lambda tokens: self.eval_forward(tokens)[0])
if self.transform_forward:
self.state_sharding = self.get_state_sharding(init_data)
rank_logger.info(f"State sharding type: {type(self.state_sharding)}")
self.init_fn = pjit.pjit(self.init, out_shardings=self.state_sharding)
def init(self, rng: jax.Array, data) -> TrainingState:
assert self.transform_forward
rng, init_rng = jax.random.split(rng)
params = self.forward.init(init_rng, data["inputs"])
return TrainingState(params=params)
def get_state_sharding(self, init_data):
assert self.transform_forward
rng = jax.random.PRNGKey(self.rng_seed)
rank_logger.info(f"partition rules: {self.model.partition_rules}")
with self.mesh:
shapes = jax.eval_shape(self.init, rng, init_data)
sharding = jax.tree_util.tree_map_with_path(
apply_rules(self.model.partition_rules()),
shapes,
)
return sharding
def load_or_init(
self,
init_data: Any,
from_checkpoint: bool = True,
init_fn: Optional[Callable] = None,
):
rng = jax.random.PRNGKey(self.rng_seed)
if not self.checkpoint_path or not from_checkpoint:
rank_logger.info("Initializing model...")
with self.mesh:
if init_fn is not None:
state = init_fn(rng, init_data)
else:
assert self.transform_forward
state = self.init_fn(rng, init_data)
rank_logger.info("Model state is newly initialized.")
else:
with self.mesh:
if init_fn:
state_shapes = jax.eval_shape(init_fn, rng, init_data)
else:
assert self.transform_forward
state_shapes = jax.eval_shape(self.init_fn, rng, init_data)
init_state = None
state = xai_checkpoint.restore(
checkpoint_path=self.checkpoint_path,
state_shapes=state_shapes,
mesh=self.mesh,
between_hosts_config=self.between_hosts_config,
state_sharding=self.state_sharding,
init_state=init_state,
params_only=True,
)
del init_state
return state
@dataclass
class Request:
prompt: str
temperature: float
nucleus_p: float
rng_seed: int
max_len: int
@dataclass
class InferenceRunner:
name: str
2024-08-08 02:54:48 +03:00
runner: ModelRunner
2024-03-15 01:03:58 +03:00
load: str
tokenizer_path: str = "/tmp/xai_data/tokenizer.model"
local_mesh_config: Tuple[int, int] = (1, 1)
between_hosts_config: Tuple[int, int] = (1, 1)
2024-08-08 02:54:48 +03:00
pad_sizes: Tuple[int] = (1024,)
2024-03-15 01:03:58 +03:00
2024-08-08 02:54:48 +03:00
def get_pad_bucket(self, size: int) -> int:
2024-03-15 01:03:58 +03:00
i = bisect.bisect_left(self.pad_sizes, size)
return self.pad_sizes[min(i, len(self.pad_sizes) - 1)]
def initialize(self):
runner = self.runner
self.runner.transform_forward = True
dummy_data = dict(
2024-08-08 02:54:48 +03:00
inputs=np.zeros((1, self.get_pad_bucket(512)), dtype=np.int32),
2024-03-15 01:03:58 +03:00
)
2024-08-08 02:54:48 +03:00
state = runner.load_or_init(
2024-03-15 01:03:58 +03:00
dummy_data,
2024-08-08 02:54:48 +03:00
from_checkpoint=False,
2024-03-15 01:03:58 +03:00
)
2024-08-08 02:54:48 +03:00
runner.params = state.params
2024-03-15 01:03:58 +03:00
self.tokenizer = sentencepiece.SentencePieceProcessor(model_file=self.tokenizer_path)
2024-08-08 02:54:48 +03:00
def text_to_token_ids(text):
ids = self.tokenizer.encode(text, out_type=int)
return ids
2024-03-15 01:03:58 +03:00
2024-08-08 02:54:48 +03:00
self.text_to_token_ids = text_to_token_ids
2024-03-15 01:03:58 +03:00
2024-08-08 02:54:48 +03:00
def predict(self, request: Request) -> str:
rng = jax.random.PRNGKey(request.rng_seed)
token_ids = self.text_to_token_ids(request.prompt)
rng, gen_rng = jax.random.split(rng)
2024-03-15 01:03:58 +03:00
2024-08-08 02:54:48 +03:00
inputs = np.array(token_ids, dtype=np.int32)[np.newaxis, :]
2024-03-15 01:03:58 +03:00
2024-08-08 02:54:48 +03:00
token_ids = jnp.array(inputs)
state = self.runner.params
2024-03-15 01:03:58 +03:00
2024-08-08 02:54:48 +03:00
settings = SampleSettings(
temperature=jnp.array([request.temperature]),
nucleus_p=jnp.array([request.nucleus_p]),
mask=jnp.ones(token_ids.shape, dtype=bool),
active=jnp.ones(token_ids.shape, dtype=bool),
2024-03-15 01:03:58 +03:00
)
2024-08-08 02:54:48 +03:00
for _ in range(request.max_len):
lm_outputs = self.runner.eval_forward(token_ids)
sample_output = sample_token(gen_rng, lm_outputs, settings)
new_token = sample_output.token_id
token_ids = jnp.concatenate([token_ids, new_token], axis=-1)
if jnp.argmax(new_token) == 0:
break
2024-03-15 01:03:58 +03:00
2024-08-08 02:54:48 +03:00
return self.tokenizer.decode(token_ids.squeeze())
2024-03-15 01:03:58 +03:00
2024-08-08 02:54:48 +03:00
def main():
runner = ModelRunner(
model=LanguageModelConfig(),
checkpoint_path="path_to_checkpoint",
2024-03-15 01:03:58 +03:00
)
2024-08-08 02:54:48 +03:00
inference_runner = InferenceRunner(
name="inference",
runner=runner,
load="path_to_load",
tokenizer_path="path_to_tokenizer_model",
local_mesh_config=(1, 1),
between_hosts_config=(1, 1),
)
inference_runner.initialize()
request = Request(
prompt="Sample text",
temperature=0.7,
nucleus_p=0.9,
2024-03-15 01:03:58 +03:00
rng_seed=42,
2024-08-08 02:54:48 +03:00
max_len=100,
2024-03-15 01:03:58 +03:00
)
2024-08-08 02:54:48 +03:00
result = inference_runner.predict(request)
print(result)
if __name__ == "__main__":
main()