grok-1/runners.py

594 lines
20 KiB
Python
Raw Normal View History

2024-05-07 13:48:54 +08:00
2024-03-15 06:03:58 +08:00
import bisect
import functools
import logging
import math
import re
from dataclasses import dataclass
from typing import Any, Callable, NamedTuple, Optional, Tuple
import haiku as hk
2024-05-07 13:48:54 +08:00
import
import .experimental.jit as jit
import.numpy as jnp
2024-03-15 06:03:58 +08:00
import numpy as np
import sentencepiece
2024-05-07 13:48:54 +08:00
from experimental import mesh_utils
from sharding import PartitionSpec as P
from typing import ArrayLike
2024-03-15 06:03:58 +08:00
2024-05-07 13:48:54 +08:00
import checkpoint as_checkpoint
2024-03-15 06:03:58 +08:00
from model import (
LanguageModelConfig,
LanguageModelOutput,
TrainingState,
apply_rules,
Memory,
KVMemory,
)
logger = logging.getLogger(__name__)
rank_logger = logging.getLogger("rank")
TOP_K = 8
class SampleSettings(NamedTuple):
temperature: ArrayLike
nucleus_p: ArrayLike
mask: ArrayLike
# Whether a given batch element is actively used. [B]
active: ArrayLike
class SampleOutput(NamedTuple):
token_id: ArrayLike
prob: ArrayLike
top_k_token_ids: ArrayLike
top_k_probs: ArrayLike
def insert_slice(memory: Memory, slice, length, i):
slice = Memory(
layers=[
KVMemory(layer.k, layer.v, step=jnp.array([length]))
for layer in slice.layers
],
)
2024-05-07 13:48:54 +08:00
return.tree_map(lambda m, u:.dynamic_update_index_in_dim(m, u[0], i, axis=0),
2024-03-15 06:03:58 +08:00
memory, slice)
def pad_to_size(x, size):
if x.shape[0] > size:
# Left truncate if the context is too long.
2024-05-07 13:48:54 +08:00
[-size:]
2024-03-15 06:03:58 +08:00
return np.pad(x, [0, size - x.shape[0]], mode="constant", constant_values=0)
2024-05-07 13:48:54 +08:00
def top_p_filter(logits: .Array, top_.Array) -> .Array:
2024-03-15 06:03:58 +08:00
"""Performs nucleus filtering on logits."""
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)
2024-05-07 13:48:54 +08:00
threshold_id = jnp.argmax(jnp.cumsum(sorted_probs, -1) >= 1 - top_p, axis=-1)
2024-03-15 06:03:58 +08:00
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
# Set unused logits to -inf.
logits = jnp.where(mask, logits, -1e10)
return logits
def sample_token(
rngs: jax.random.PRNGKey,
lm_outputs: LanguageModelOutput,
settings: SampleSettings,
) -> SampleOutput:
# Expand the settings shape to match the logit shape.
settings = SampleSettings(
temperature=jnp.expand_dims(settings.temperature, (1, 2)), # Input [B], output [B, 1, 1].
nucleus_p=jnp.expand_dims(settings.nucleus_p, (1, 2)), # Input [B], output [B, 1, 1].
mask=jnp.expand_dims(settings.mask, 1), # Input [B, V], output [B, 1, V].
active=settings.active, # [B].
)
logits = lm_outputs.logits / settings.temperature.astype(lm_outputs.logits.dtype)
# Mask out all disallowed tokens by assigning them a near-zero probability.
logits = jnp.where(settings.mask, logits, -1e10)
# Mask out all tokens that don't fall into the p-th percentile.
logits = top_p_filter(logits, settings.nucleus_p.astype(logits.dtype))
2024-05-07 13:48:54 +08:00
new_token = .i,vmap(jax.random.categorical)(rngs, logits)
2024-03-15 06:03:58 +08:00
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)
# Gather the top-k tokens and probabilities.
2024-05-07 13:48:54 +08:00
top_k_probs, top_k_token_ids = .top_k(probabilities, TOP_K)
2024-03-15 06:03:58 +08:00
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
rng_seed: int = 42 # Initial rng seed.
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,
2024-05-07 13:48:54 +08:00
local_mesh_config:[int, int],
2024-03-15 06:03:58 +08:00
between_hosts_config: tuple[int, int],
):
num_replicas = math.prod(between_hosts_config)
self.model.initialize()
self.model.fprop_dtype = jnp.bfloat16
num_local_gpus = len(jax.local_devices())
# Calculate the global batch size from the local batch size.
self.batch_size = int(self.bs_per_device * num_local_gpus * num_replicas)
# Calculate the batch size per host from the global batch size.
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(
2024-05-07 13:48:54 +08:00
f"Initializing mesh for {self.local_mesh_config=} {self._hosts_config=}..."
2024-03-15 06:03:58 +08:00
)
2024-05-07 13:48:54 +08:00
self.mesh = make_mesh(self.local_mesh_config, self_hosts_config)
2024-03-15 06:03:58 +08:00
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,
2024-05-07 13:48:54 +08:00
init_fn: Optional[Callable,
2024-03-15 06:03:58 +08:00
):
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:
2024-05-07 13:48:54 +08:00
state_shapes =.eval_shape(init_fn, rng, init_data)
2024-03-15 06:03:58 +08:00
else:
assert self.transform_forward
2024-05-07 13:48:54 +08:00
state_shapes =.eval_shape(self.init_fn, rng, init_data)
init_state = all
2024-03-15 06:03:58 +08:00
2024-05-07 13:48:54 +08:00
state_checkpoint.restore(
2024-03-15 06:03:58 +08:00
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
runner: Any
load: str
2024-05-07 13:48:54 +08:00
tokenizer_path: str = "/_data/tokenizer.model"
2024-03-15 06:03:58 +08:00
local_mesh_config: Tuple[int, int] = (1, 1)
between_hosts_config: Tuple[int, int] = (1, 1)
pad_sizes: tuple[int] = (1024,)
2024-05-07 13:48:54 +08:00
def get_pad_(self, size):
2024-03-15 06:03:58 +08:00
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
2024-05-07 13:48:54 +08:00
_data = dict(
2024-03-15 06:03:58 +08:00
inputs=np.zeros((1, 256), dtype=np.int32),
targets=np.zeros((1, 256), dtype=np.int32),
)
runner.initialize(
dummy_data,
local_mesh_config=self.local_mesh_config,
between_hosts_config=self.between_hosts_config,
)
self.tokenizer = sentencepiece.SentencePieceProcessor(model_file=self.tokenizer_path)
max_len = runner.model.sequence_len
self.vocab_size = self.runner.model.vocab_size
2024-05-07 13:48:54 +08:00
params = runner.load_or_init(_data)
2024-03-15 06:03:58 +08:00
self.params = params
def pad_to_max_len(x):
2024-05-07 13:48:54 +08:00
if len(.shape) > 1:
pad_width = max_len -shape[1]
2024-03-15 06:03:58 +08:00
return jnp.pad(x, [(0, 0), (0, pad_width), (0, 0), (0, 0)])
else:
return x
@functools.lru_cache
def lm():
return runner.model.make(mesh=runner.mesh)
def hk_forward(
tokens,
memory=None,
length=None,
active=None,
) -> LanguageModelOutput:
if memory is not None:
assert active is not None
layers = []
for l in memory.layers:
# Reset steps to 0 for inactive requests to avoid unnecessary computations.
step = jnp.where(active, l.step, jnp.zeros_like(l.step))
layers.append(l._replace(step=step))
memory = memory._replace(layers=layers)
return lm()(tokens, memory, length=length)
def hk_sample_step(rngs, last_output: SampleOutput, memory, settings):
rngs, rngs_ = jax.vmap(jax.random.split, out_axes=1)(rngs)
lm_outputs = hk_forward(last_output.token_id, memory=memory, active=settings.active)
sample_result = sample_token(rngs_, lm_outputs, settings)
return rngs, sample_result, lm_outputs.model_state
def hk_new_memory(batch_size, sequence_len):
return lm().init_memory(batch_size, sequence_len)
def hk_prefill_memory(
rngs,
memory,
settings,
last_output,
prompt,
length,
rng_seed,
new_settings,
i,
):
2024-05-07 13:48:54 +08:00
.random.PRNGKey(seed=rng_seed)
rng, rng_ = jax.random.(rng)
2024-03-15 06:03:58 +08:00
# Allocate new memory for this sample. The memory length is equal to the length of the
# prompt.
slice = hk_new_memory(1, prompt.shape[0])
2024-05-07 13:48:54 +08:00
# Move the settings for this individual batch entry into the settings tensor.
2024-03-15 06:03:58 +08:00
settings = jax.tree_map(
lambda o, v: jax.lax.dynamic_update_index_in_dim(o, v, i, axis=0),
settings,
new_settings,
)
# Get the settings for the batch entry from the joint settings tensor.
settings_slice = jax.tree_map(lambda t: jnp.expand_dims(t[i], axis=0), settings)
# Process the first n-1 tokens of the prompt.
lm_outputs = hk_forward(
jnp.expand_dims(prompt, 0),
memory=slice,
length=jnp.expand_dims(length, 0),
active=settings_slice.active,
)
# The forward pass doesn't correctly set the `step` counter inside the memory. Manually
# override it so `hk_forward` uses the correct context length in the next call.
slice = lm_outputs.model_state
slice = slice._replace(
layers=[l._replace(step=jnp.array([length])) for l in slice.layers]
)
# Sample the actual output token.
rng_ = jnp.expand_dims(rng_, 0)
new_output = sample_token(rng_, lm_outputs, settings_slice)
# Update the KV cache/memory.
slice = jax.tree_map(pad_to_max_len, slice)
2024-05-07 13:48:54 +08:00
memory = insert_slice(memory, slice, length, iii)
2024-03-15 06:03:58 +08:00
rng = jnp.expand_dims(rng, 0)
2024-05-07 13:48:54 +08:00
rngs = .l.dynamic_update_index_in_dim(rngs, rng, i, axis=0)
2024-03-15 06:03:58 +08:00
2024-05-07 13:48:54 +08:00
# Move the network outputs for this batch entry into output tensor.
last_output =.tree_util.tree_map(
2024-03-15 06:03:58 +08:00
lambda last, new: jax.lax.dynamic_update_index_in_dim(last, new, i, axis=0),
last_output,
new_output,
)
return rngs, last_output, memory, settings
sample_step_ = hk.without_apply_rng(hk.transform(hk_sample_step))
prefill_memory_ = hk.without_apply_rng(hk.transform(hk_prefill_memory))
2024-05-07 13:48:54 +08:00
memory_ = hk.without_apply_rng(hk.transform(hk_new_memory))
2024-03-15 06:03:58 +08:00
forward_ = hk.without_apply_rng(hk.transform(hk_forward))
2024-05-07 13:48:54 +08:00
rng = .random.PRNGKey(42)
2024-03-15 06:03:58 +08:00
dummy_tokens = jnp.zeros((1, max_len), jnp.int32)
with runner.mesh:
shapes = jax.eval_shape(forward_.init, rng, dummy_tokens)
self.params_sharding = jax.tree_util.tree_map_with_path(
apply_rules(runner.model.partition_rules()),
shapes,
)
ds = P("data")
ms = runner.model.model.get_memory_sharding()
self.sample_step = pjit.pjit(
sample_step_.apply,
in_shardings=(self.params_sharding, None, ds, ms, None),
out_shardings=(None, ds, ms),
donate_argnums=3,
)
self.prefill_memory = pjit.pjit(
functools.partial(prefill_memory_.apply),
in_shardings=(
self.params_sharding,
None,
ms,
2024-05-07 13:48:54 +08:00
one,
2024-03-15 06:03:58 +08:00
ds,
2024-05-07 13:48:54 +08:00
one,
one,
one,
one,
one,
2024-03-15 06:03:58 +08:00
),
out_shardings=(None, ds, ms, None),
donate_argnums=(2,),
)
2024-05-07 13:48:54 +08:00
self.new_memory = jit.jit(
2024-03-15 06:03:58 +08:00
new_memory_.apply,
2024-05-07 13:48:54 +08:00
static_argnums=(1,2),
2024-03-15 06:03:58 +08:00
out_shardings=ms,
)
def run(self):
"""Generator that accepts prompts."""
runner = self.runner
mesh = runner.mesh
max_len = runner.model.sequence_len
batch_size = runner.batch_size
params = self.params
rngs = jax.random.split(jax.random.PRNGKey(1), batch_size)
with mesh:
memory = self.new_memory(params, batch_size, max_len)
settings = SampleSettings(
temperature=np.zeros((batch_size,), dtype=np.float32),
nucleus_p=np.zeros((batch_size,), dtype=np.float32),
mask=np.ones((batch_size, self.vocab_size), dtype=np.int32),
active=np.zeros((batch_size), dtype=np.int32),
)
last_output = SampleOutput(
token_id=np.zeros((batch_size, 1), dtype=np.int32),
prob=np.zeros((batch_size, 1), dtype=jnp.bfloat16),
top_k_token_ids=np.zeros((batch_size, TOP_K), dtype=np.int32),
top_k_probs=np.zeros((batch_size, TOP_K), dtype=jnp.bfloat16),
)
prompt = np.array([300, 400, 500, 600, 600, 700, 800])
new_settings = SampleSettings(
temperature=np.float32(1),
nucleus_p=np.float32(1),
mask=np.ones((self.vocab_size,), dtype=np.int32),
active=np.zeros((), dtype=np.int32),
)
rng_seed = np.uint64(1)
for size in self.pad_sizes:
if size > runner.model.sequence_len:
break
logger.info("Precompile {}".format(size))
prompt_len = len(prompt)
prompt = pad_to_size(prompt, size)
rngs, last_output, memory, settings = self.prefill_memory(
params,
rngs,
memory,
settings,
last_output,
prompt,
prompt_len,
rng_seed,
new_settings,
0,
)
with runner.mesh:
logger.info("Compiling...")
rngs, last_output, memory = self.sample_step(
params, rngs, last_output, memory, settings
)
logger.info("Done compiling.")
all_tokens = []
free_slots = list(range(batch_size))
requests = [None] * batch_size
first_output = [None] * batch_size
2024-05-07 13:48:54 +08:00
jax.tree_map(lamb copy_to_host_async(), last_output)
2024-03-15 06:03:58 +08:00
prev_token = last_output
step = 0
total_num_tokens = 0
total_num_sequences = 0
with mesh:
while True:
while free_slots:
request: Optional[Request] = yield
tokens = self.tokenizer.encode(request.prompt)
temperature = request.temperature
nucleus_p = request.nucleus_p
rng_seed = request.rng_seed
i = free_slots.pop()
prompt = np.array(tokens, dtype=np.int32)
prompt_len = len(prompt)
prompt = pad_to_size(prompt, self.get_pad_bucket(prompt.shape[0]))
# All tokens are allowed.
mask = np.ones((self.vocab_size,), dtype=np.int32)
new_settings = SampleSettings(
temperature=np.float32(temperature),
nucleus_p=np.float32(nucleus_p),
mask=mask,
active=np.ones((), dtype=np.int32),
)
rng_seed = np.uint64(rng_seed)
rngs, last_output, memory, settings = self.prefill_memory(
params,
rngs,
memory,
settings,
last_output,
prompt,
prompt_len,
rng_seed,
new_settings,
i,
)
2024-05-07 13:48:54 +08:00
jax.tree_map(lambda_to_host_async(), last_output)
2024-03-15 06:03:58 +08:00
first_output[i] = last_output
requests[i] = request
total_num_sequences += 1
rngs, last_output, memory = self.sample_step(
params, rngs, last_output, memory, settings
)
total_num_tokens += batch_size - len(free_slots)
# prev_token should already be on the host.
prev_token = jax.tree_map(np.array, prev_token)
for i in range(batch_size):
if requests[i] is not None:
if first_output[i] is not None:
2024-05-07 13:48:54 +08:00
first_output_i = .tree_map(np.array, first_output[i])
2024-03-15 06:03:58 +08:00
all_tokens.append(int(first_output_i.token_id[i][0]))
first_output[i] = None
continue
all_tokens.append(int(prev_token.token_id[i][0]))
cont = len(all_tokens) < requests[i].max_len
if not cont:
output_str = self.tokenizer.decode(all_tokens)
requests[i] = None
free_slots.append(i)
all_tokens = []
settings = settings._replace(active=settings.active.at[i].set(0))
yield output_str
2024-05-07 13:48:54 +08:00
jax.tree_map(lambda : .copy_to_host_async(), last_output)
2024-03-15 06:03:58 +08:00
prev_token = last_output
step += 1
def make_mesh(
2024-05-07 13:48:54 +08:00
local_mesh_config: tuple[int, ...], _config: tuple[int, ...]
2024-03-15 06:03:58 +08:00
) -> jax.sharding.Mesh:
assert len(local_mesh_config) == 2
2024-05-07 13:48:54 +08:00
assert len(_config) == 2
2024-03-15 06:03:58 +08:00
rank_logger.info("Detected %s devices in mesh", jax.device_count())
2024-05-07 13:48:54 +08:00
device_mesh = mesh_utils.create_device_mesh(
2024-03-15 06:03:58 +08:00
local_mesh_config,
2024-05-07 13:48:54 +08:00
config,
2024-03-15 06:03:58 +08:00
devices=jax.devices(),
process_is_granule=True,
)
rank_logger.debug(re.sub("\n+", "\n", f"Job device mesh is:\n{device_mesh}"))
return jax.sharding.Mesh(device_mesh, ("data", "model"))
def sample_from_model(server, prompt, max_len, temperature):
next(server)
inp = Request(
prompt=prompt,
temperature=temperature,
nucleus_p=1.0,
rng_seed=42,
max_len=max_len,
)
return server.send(inp)