berkus
/
ANE
mirror of https://github.com/maderix/ANE.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Fix dashboard text generation: add KV cache for proper autoregressive attention

This commit is contained in:
maderix 2026-03-05 08:14:21 -08:00
parent 19da850fca
commit 06535fc5be
1 changed files with 14 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
training/dashboard.py
View File

@ -162,12 +162,17 @@ def generate_text(W, max_tokens=64, temperature=0.8):
freq = 1.0 / (10000.0 ** (2.0 * i / HD))
freqs[pos, i] = pos * freq
# KV cache: per-layer, per-head arrays
k_cache = [[np.zeros((0, HD), dtype=np.float32) for _ in range(HEADS)] for _ in range(NLAYERS)]
v_cache = [[np.zeros((0, HD), dtype=np.float32) for _ in range(HEADS)] for _ in range(NLAYERS)]
for step in range(max_tokens):
seq_len = len(tokens)
if seq_len > SEQ:
break
x = W['embed'][tokens[-1]].copy()
pos = seq_len - 1
for L in range(NLAYERS):
# RMSNorm + QKV
@ -177,7 +182,6 @@ def generate_text(W, max_tokens=64, temperature=0.8):
v = W[f'Wv{L}'] @ xn
# RoPE
pos = seq_len - 1
for h in range(HEADS):
for i in range(HD // 2):
freq = freqs[pos, i]
@ -189,14 +193,18 @@ def generate_text(W, max_tokens=64, temperature=0.8):
k[h * HD + 2 * i] = ki * cos_v - ki1 * sin_v
k[h * HD + 2 * i + 1] = ki * sin_v + ki1 * cos_v
# Attention (single token)
# Append to KV cache and compute attention
o = np.zeros(DIM, dtype=np.float32)
for h in range(HEADS):
qh = q[h * HD:(h + 1) * HD]
kh = k[h * HD:(h + 1) * HD]
vh = v[h * HD:(h + 1) * HD]
score = np.dot(qh, kh) / math.sqrt(HD)
o[h * HD:(h + 1) * HD] = vh
kh = k[h * HD:(h + 1) * HD].reshape(1, HD)
vh = v[h * HD:(h + 1) * HD].reshape(1, HD)
k_cache[L][h] = np.vstack([k_cache[L][h], kh])
v_cache[L][h] = np.vstack([v_cache[L][h], vh])
# scores: (1, HD) @ (HD, seq_len) -> (seq_len,)
scores = k_cache[L][h] @ qh / math.sqrt(HD)
attn = softmax(scores)
o[h * HD:(h + 1) * HD] = attn @ v_cache[L][h]
# Residual + output projection
x2 = x + W[f'Wo{L}'] @ o