yujunjun/LLMs-from-scratch
1
0
Fork 0
mirror of https://github.com/rasbt/LLMs-from-scratch.git synced 2025-11-03 11:20:49 +00:00
Code Issues Packages Projects Releases Wiki Activity

Simplify KV cache usage (#728)

Browse Source 
* Simplify KV cache usage

* Swap mark text with ghostwriter
This commit is contained in:
Sebastian Raschka 2025-07-08 12:56:55 -05:00 committed by GitHub
parent b5bd8d2de2
commit 90c824506c
4 changed files with 31 additions and 39 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
README.md
View File

@ -40,7 +40,7 @@ git clone --depth 1 https://github.com/rasbt/LLMs-from-scratch.git
# Table of Contents
Please note that this `README.md` file is a Markdown (`.md`) file. If you have downloaded this code bundle from the Manning website and are viewing it on your local computer, I recommend using a Markdown editor or previewer for proper viewing. If you haven't installed a Markdown editor yet, [MarkText](https://www.marktext.cc) is a good free option.
Please note that this `README.md` file is a Markdown (`.md`) file. If you have downloaded this code bundle from the Manning website and are viewing it on your local computer, I recommend using a Markdown editor or previewer for proper viewing. If you haven't installed a Markdown editor yet, [Ghostwriter](https://ghostwriter.kde.org) is a good free option.
You can alternatively view this and other files on GitHub at [https://github.com/rasbt/LLMs-from-scratch](https://github.com/rasbt/LLMs-from-scratch) in your browser, which renders Markdown automatically.

8
pkg/llms_from_scratch/kv_cache/generate.py
View File

@ -10,20 +10,20 @@ import torch
def generate_text_simple(model, idx, max_new_tokens, context_size=None, use_cache=True):
model.eval()
ctx_len = context_size or model.cfg["context_length"]
cache = KVCache(n_layers=model.cfg["n_layers"]) if use_cache else None
with torch.no_grad():
if use_cache:
cache = KVCache(n_layers=model.cfg["n_layers"])
model.reset_kv_cache()
logits = model(idx[:, -ctx_len:], use_cache=True, cache=cache)
logits = model(idx[:, -ctx_len:], cache=cache)
for _ in range(max_new_tokens):
next_idx = logits[:, -1].argmax(dim=-1, keepdim=True)
idx = torch.cat([idx, next_idx], dim=1)
logits = model(next_idx, use_cache=True, cache=cache)
logits = model(next_idx, cache=cache)
else:
for _ in range(max_new_tokens):
logits = model(idx[:, -ctx_len:], use_cache=False)
logits = model(idx[:, -ctx_len:], cache=None)
next_idx = logits[:, -1].argmax(dim=-1, keepdim=True)
idx = torch.cat([idx, next_idx], dim=1)

32
pkg/llms_from_scratch/kv_cache/llama3.py
View File

@ -77,12 +77,12 @@ class Llama3Model(nn.Module):
self.cfg = cfg
self.current_pos = 0 # Track current position in KV cache
def forward(self, in_idx, use_cache=False, cache=None):
def forward(self, in_idx, cache=None):
tok_embeds = self.tok_emb(in_idx)
x = tok_embeds
num_tokens = x.shape[1]
if use_cache:
if cache is not None:
pos_start = self.current_pos
pos_end = pos_start + num_tokens
self.current_pos = pos_end
@ -101,10 +101,9 @@ class Llama3Model(nn.Module):
for i, block in enumerate(self.trf_blocks):
blk_cache = cache.get(i) if cache else None
x, new_blk_cache = block(x, mask, self.cos, self.sin,
use_cache=use_cache,
start_pos=pos_start,
cache=blk_cache)
if cache:
if cache is not None:
cache.update(i, new_blk_cache)
next_cache.append(new_blk_cache)
@ -130,11 +129,11 @@ class TransformerBlock(nn.Module):
self.norm1 = nn.RMSNorm(cfg["emb_dim"], eps=1e-5, dtype=cfg["dtype"])
self.norm2 = nn.RMSNorm(cfg["emb_dim"], eps=1e-5, dtype=cfg["dtype"])
def forward(self, x, mask, cos, sin, use_cache=False, start_pos=0, cache=None):
def forward(self, x, mask, cos, sin, start_pos=0, cache=None):
# Shortcut connection for attention block
shortcut = x
x = self.norm1(x)
x, next_cache = self.att(x, mask, cos, sin, use_cache=use_cache, start_pos=start_pos, cache=cache) # Shape [batch_size, num_tokens, emb_size]
x, next_cache = self.att(x, mask, cos, sin, start_pos=start_pos, cache=cache) # Shape [batch_size, num_tokens, emb_size]
x = x + shortcut # Add the original input back
# Shortcut connection for feed-forward block
@ -180,7 +179,7 @@ class GroupedQueryAttention(nn.Module):
self.W_query = nn.Linear(d_in, d_out, bias=False, dtype=dtype)
self.out_proj = nn.Linear(d_out, d_out, bias=False, dtype=dtype)
def forward(self, x, mask, cos, sin, use_cache=False, start_pos=0, cache=None):
def forward(self, x, mask, cos, sin, start_pos=0, cache=None):
b, num_tokens, _ = x.shape
# Apply projections
@ -197,18 +196,15 @@ class GroupedQueryAttention(nn.Module):
queries = apply_rope(queries, cos, sin, offset=start_pos)
keys_new = apply_rope(keys_new, cos, sin, offset=start_pos)
if use_cache:
if cache is None:
keys = keys_new
values = values_new
else:
prev_k, prev_v = cache
keys = torch.cat([prev_k, keys_new], dim=2)
values = torch.cat([prev_v, values_new], dim=2)
if cache is not None:
prev_k, prev_v = cache
keys = torch.cat([prev_k, keys_new], dim=2)
values = torch.cat([prev_v, values_new], dim=2)
next_cache = (keys, values)
else:
start_pos = 0 # reset RoPE
keys, values = keys_new, values_new
next_cache = None
next_cache = (keys, values)
# Expand keys and values to match the number of heads
# Shape: (b, num_heads, num_tokens, head_dim)
@ -226,7 +222,7 @@ class GroupedQueryAttention(nn.Module):
attn_scores = queries @ keys.transpose(2, 3) # Dot product for each head
# Use the mask to fill attention scores
attn_scores = attn_scores.masked_fill(mask[:num_tokens, :num_tokens], -torch.inf)
attn_scores = attn_scores.masked_fill(mask, -torch.inf)
attn_weights = torch.softmax(attn_scores / keys.shape[-1]**0.5, dim=-1)
assert keys.shape[-1] == self.head_dim
@ -286,7 +282,7 @@ def compute_rope_params(head_dim, theta_base=10_000, context_length=4096, freq_c
return cos, sin
def apply_rope(x, cos, sin, offset=9):
def apply_rope(x, cos, sin, offset=0):
# x: (batch_size, num_heads, seq_len, head_dim)
batch_size, num_heads, seq_len, head_dim = x.shape
assert head_dim % 2 == 0, "Head dimension must be even"

28
pkg/llms_from_scratch/kv_cache/qwen3.py
View File

@ -44,13 +44,13 @@ class Qwen3Model(nn.Module):
self.cfg = cfg
self.current_pos = 0 # Track current position in KV cache
def forward(self, in_idx, use_cache=False, cache=None):
def forward(self, in_idx, cache=None):
# Forward pass
tok_embeds = self.tok_emb(in_idx)
x = tok_embeds
num_tokens = x.shape[1]
if use_cache:
if cache is not None:
pos_start = self.current_pos
pos_end = pos_start + num_tokens
self.current_pos = pos_end
@ -69,10 +69,9 @@ class Qwen3Model(nn.Module):
for i, block in enumerate(self.trf_blocks):
blk_cache = cache.get(i) if cache else None
x, new_blk_cache = block(x, mask, self.cos, self.sin,
use_cache=use_cache,
start_pos=pos_start,
cache=blk_cache)
if cache:
if cache is not None:
cache.update(i, new_blk_cache)
next_cache.append(new_blk_cache)
@ -99,11 +98,11 @@ class TransformerBlock(nn.Module):
self.norm1 = RMSNorm(cfg["emb_dim"], eps=1e-6)
self.norm2 = RMSNorm(cfg["emb_dim"], eps=1e-6)
def forward(self, x, mask, cos, sin, use_cache=False, start_pos=0, cache=None):
def forward(self, x, mask, cos, sin, start_pos=0, cache=None):
# Shortcut connection for attention block
shortcut = x
x = self.norm1(x)
x, next_cache = self.att(x, mask, cos, sin, use_cache=use_cache, start_pos=start_pos, cache=cache) # Shape [batch_size, num_tokens, emb_size]
x, next_cache = self.att(x, mask, cos, sin, start_pos=start_pos, cache=cache) # Shape [batch_size, num_tokens, emb_size]
x = x + shortcut # Add the original input back
# Shortcut connection for feed-forward block
@ -159,7 +158,7 @@ class GroupedQueryAttention(nn.Module):
else:
self.q_norm = self.k_norm = None
def forward(self, x, mask, cos, sin, use_cache=False, start_pos=0, cache=None):
def forward(self, x, mask, cos, sin, start_pos=0, cache=None):
b, num_tokens, _ = x.shape
# Apply projections
@ -182,18 +181,15 @@ class GroupedQueryAttention(nn.Module):
queries = apply_rope(queries, cos, sin, offset=start_pos)
keys_new = apply_rope(keys_new, cos, sin, offset=start_pos)
if use_cache:
if cache is None:
keys = keys_new
values = values_new
else:
prev_k, prev_v = cache
keys = torch.cat([prev_k, keys_new], dim=2)
values = torch.cat([prev_v, values_new], dim=2)
if cache is not None:
prev_k, prev_v = cache
keys = torch.cat([prev_k, keys_new], dim=2)
values = torch.cat([prev_v, values_new], dim=2)
next_cache = (keys, values)
else:
start_pos = 0 # reset RoPE
keys, values = keys_new, values_new
next_cache = None
next_cache = (keys, values)
# Expand K and V to match number of heads
keys = keys.repeat_interleave(self.group_size, dim=1)