Dataset Viewer
Auto-converted to Parquet Duplicate
layer
int64
2
23
num_tokens
int64
1k
1k
num_experts
int64
32
32
top_k
int64
4
4
stats
dict
all_raw_logits
listlengths
1k
1k
per_token
listlengths
1k
1k
2
1,000
32
4
{"mean_top1_logit":0.8828,"std_top1_logit":0.4785,"mean_topk_logit":-0.0491,"std_logits":0.8086,"std(...TRUNCATED)
[[-2.265625,-1.8828125,-2.359375,-3.09375,-1.75,-2.21875,-3.15625,-2.15625,-3.515625,-2.703125,-1.92(...TRUNCATED)
[{"token_idx":0,"token":"We","top1_logit":-0.6133,"selected_experts":[21,25,18,14],"gate_weights":[0(...TRUNCATED)
3
1,000
32
4
{"mean_top1_logit":0.5273,"std_top1_logit":0.4766,"mean_topk_logit":-0.4629,"std_logits":0.7695,"std(...TRUNCATED)
[[-2.671875,-2.1875,-3.25,-2.578125,-2.609375,-0.275390625,-3.03125,-2.515625,-1.40625,-1.671875,-1.(...TRUNCATED)
[{"token_idx":0,"token":"We","top1_logit":-0.2754,"selected_experts":[5,29,30,10],"gate_weights":[0.(...TRUNCATED)
4
1,000
32
4
{"mean_top1_logit":0.6328,"std_top1_logit":0.7344,"mean_topk_logit":-0.4375,"std_logits":0.7734,"std(...TRUNCATED)
[[-1.8515625,-2.171875,-1.0390625,-1.078125,-2.234375,-2.15625,-1.4765625,-2.296875,-2.203125,-1.648(...TRUNCATED)
[{"token_idx":0,"token":"We","top1_logit":-0.7109,"selected_experts":[26,2,27,15],"gate_weights":[0.(...TRUNCATED)
5
1,000
32
4
{"mean_top1_logit":0.4297,"std_top1_logit":0.6016,"mean_topk_logit":-0.4004,"std_logits":0.7539,"std(...TRUNCATED)
[[-1.859375,-1.9375,-1.8359375,-1.4140625,-1.265625,-2.0625,-2.015625,-1.953125,-2.625,-2.75,-1.7734(...TRUNCATED)
[{"token_idx":0,"token":"We","top1_logit":-1.2109,"selected_experts":[17,29,3,4],"gate_weights":[0.2(...TRUNCATED)
6
1,000
32
4
{"mean_top1_logit":0.6133,"std_top1_logit":0.6172,"mean_topk_logit":-0.1709,"std_logits":0.7695,"std(...TRUNCATED)
[[-0.9140625,-3.703125,-0.43359375,-1.953125,0.4609375,-2.625,-1.484375,-2.375,-2.984375,-1.640625,-(...TRUNCATED)
[{"token_idx":0,"token":"We","top1_logit":0.4609,"selected_experts":[11,4,26,14],"gate_weights":[0.2(...TRUNCATED)
7
1,000
32
4
{"mean_top1_logit":0.4434,"std_top1_logit":0.6055,"mean_topk_logit":-0.4258,"std_logits":0.6992,"std(...TRUNCATED)
[[-2.140625,-2.1875,-1.3046875,-0.3125,-2.890625,-2.015625,-3.203125,-2.1875,-2.296875,-2.21875,-0.6(...TRUNCATED)
[{"token_idx":0,"token":"We","top1_logit":-0.3125,"selected_experts":[3,10,2,16],"gate_weights":[0.3(...TRUNCATED)
8
1,000
32
4
{"mean_top1_logit":0.4648,"std_top1_logit":0.543,"mean_topk_logit":-0.3652,"std_logits":0.7266,"std_(...TRUNCATED)
[[-2.4375,-0.8359375,-2.46875,-2.34375,-3.25,-2.515625,-2.75,-3.53125,-2.828125,-2.453125,-2.6875,-1(...TRUNCATED)
[{"token_idx":0,"token":"We","top1_logit":-0.8359,"selected_experts":[1,11,30,22],"gate_weights":[0.(...TRUNCATED)
9
1,000
32
4
{"mean_top1_logit":0.5,"std_top1_logit":0.5508,"mean_topk_logit":-0.2158,"std_logits":0.8398,"std_ga(...TRUNCATED)
[[-3.046875,-2.484375,-2.21875,-2.71875,-2.671875,-1.8828125,-2.015625,-1.390625,-2.03125,-2.421875,(...TRUNCATED)
[{"token_idx":0,"token":"We","top1_logit":-0.1943,"selected_experts":[17,11,21,7],"gate_weights":[0.(...TRUNCATED)
10
1,000
32
4
{"mean_top1_logit":0.5312,"std_top1_logit":0.5859,"mean_topk_logit":-0.1797,"std_logits":0.8203,"std(...TRUNCATED)
[[-0.67578125,-1.96875,0.609375,-1.5390625,-0.98046875,-1.5625,-1.359375,-1.4375,-1.640625,-1.9375,-(...TRUNCATED)
[{"token_idx":0,"token":"We","top1_logit":0.6094,"selected_experts":[2,30,0,21],"gate_weights":[0.37(...TRUNCATED)
11
1,000
32
4
{"mean_top1_logit":0.3809,"std_top1_logit":0.5898,"mean_topk_logit":-0.3477,"std_logits":0.7461,"std(...TRUNCATED)
[[-1.8125,-1.828125,-2.046875,-1.625,-1.59375,-2.640625,-2.015625,-1.828125,-2.15625,-3.0,-1.3515625(...TRUNCATED)
[{"token_idx":0,"token":"We","top1_logit":0.3496,"selected_experts":[26,11,20,14],"gate_weights":[0.(...TRUNCATED)
End of preview. Expand in Data Studio

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

LFM2.5-8B-A1B Router Logit Analysis

Model: LiquidAI/LFM2.5-8B-A1B

Date: 2026-07-21T20:29:36.234301Z

Tokens processed: 1000

MoE layers analyzed: 22

Experts per layer: 32

Trained top-k: 4

Files

  • router_logits.json — Full per-token, per-layer data including ALL 32 raw logits
  • router_logits_raw.pt — PyTorch dict of raw logits tensors [N, 32] per layer
  • router_summary.txt — Human-readable summary with statistics
  • README.md — This file
Downloads last month
24