zerodegress commited on
Commit
07cdc14
·
verified ·
1 Parent(s): 9abac51

Upload MANIFEST.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. MANIFEST.json +34 -0
MANIFEST.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema_version": "nanojev-mlx-quant-v1",
3
+ "source_checkpoint": "NanoJev",
4
+ "source_weights_sha256": "fff62d1412685c1714eaa386acb603f9690371fb3cc8ad03dc41319302597c28",
5
+ "bits": 4,
6
+ "group_size": 64,
7
+ "include_embed": true,
8
+ "scales_dtype": "float16",
9
+ "note": "backbone linears + embed_tokens in MLX affine 4-bit (group 64); head and norms kept in fp16",
10
+ "quantized_params_by_bucket": {
11
+ "mlp": 264241152,
12
+ "self_attn": 176160768,
13
+ "embed_tokens": 155582464
14
+ },
15
+ "relative_max_error_by_bucket": {
16
+ "embed_tokens": {
17
+ "max": 0.06698610633611679,
18
+ "mean": 0.06698610633611679
19
+ },
20
+ "mlp": {
21
+ "max": 0.12449125945568085,
22
+ "mean": 0.06786140923698743
23
+ },
24
+ "self_attn": {
25
+ "max": 0.11749649047851562,
26
+ "mean": 0.07736335305630096
27
+ }
28
+ },
29
+ "packed_weights_bytes": 335857708,
30
+ "packed_weights_sha256": "332d47eb23c81b0370f580e87e0e55ef737ba1baa206a8d300eadf9617307b62",
31
+ "dequant_fp32_bytes": 2384507132,
32
+ "dequant_fp32_sha256": "2e5bf94be4fa2264e95e4ee3ba763d6240a41faaff6f6cd643f9e1eea29fe291",
33
+ "seconds": 87.46
34
+ }