Skip to content

Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦

C

tools: rate-scale the E8 lattice ball by bit-width — int3-e8 is now a real int3 codebook, not a fixed 2-bit ball

ZacharyZcR3f239dbb922e
diff --git a/c/tools/quant_ablation.py b/c/tools/quant_ablation.py
index 689643af..dd087152 100644
--- a/c/tools/quant_ablation.py
+++ b/c/tools/quant_ablation.py
@@ -118,7 +118,7 @@ def quantize_param(w, bits, group, rot=False, e8=""):
def _grid_or_e8(x, bits, group, e8):
if e8:
- return _quant_e8(x.float(), group, ball=(e8 == "-e8"))
+ return _quant_e8(x.float(), group, bits, ball=(e8 == "-e8"))
return _quant_last_dim(x, bits, group)
@@ -169,8 +169,15 @@ def _e8_ball(y, r2=10.0):
return p
-def _quant_e8(x, group, ball):
- """Blocks of 8 along the input dim; per-group scale by MSE search over RMS multiples."""
+_E8_R2_REPORTED = set()
+def _e8_radius(bits):
+ # E8 lattice: points within |p|^2<=r2 grow ~r2^4, so +1 bit (x256 codebook) needs r2 x4.
+ # Anchor: r2=10 is the ~2^16 E8P ball (2 bits over 8 dims). Scale from there.
+ return 10.0 * (4.0 ** (bits - 2))
+
+def _quant_e8(x, group, bits, ball):
+ """Blocks of 8 along the input dim; per-group scale by MSE search over RMS multiples.
+ ball=True clamps to the rate-scaled E8 ball for `bits`; ball=False is the unbounded ideal."""
if x.shape[-1] % 8:
raise SystemExit(f"-e8 needs input dim divisible by 8 (got {x.shape[-1]})")
g = group or x.shape[-1]
@@ -183,7 +190,11 @@ def _quant_e8(x, group, ball):
for k in (0.5, 0.7, 0.9, 1.1, 1.4, 1.8, 2.4):
s = rms * k
yb = (xg / s).reshape(-1, g // 8, 8)
- p = _e8_ball(yb) if ball else _e8_nearest(yb)
+ p = _e8_ball(yb, _e8_radius(bits)) if ball else _e8_nearest(yb)
+ if ball and bits not in _E8_R2_REPORTED:
+ _E8_R2_REPORTED.add(bits)
+ import sys as _sys
+ _sys.stderr.write(f"[e8] bits={bits}: ball r2={_e8_radius(bits):.1f}\n")
out = (p.reshape(-1, g) * s)
err = (out - xg).pow(2).sum(-1, keepdim=True)
if best_err is None: