sanmonga22 commited on
Commit
989632c
·
verified ·
1 Parent(s): 205f022

Prune QHexRT HNPU bundle to runtime-minimum artifacts

Browse files

Strip non-runtime _comment metadata from 1 manifest(s) without changing runtime fields.
Audit artifacts: QHexRT/hf_hnpu_audit cleanup plan, manifest reference check, and multi-agent review.

Files changed (1) hide show
  1. v81/nv-embedcode-7b.json +39 -8
v81/nv-embedcode-7b.json CHANGED
@@ -1,17 +1,48 @@
1
  {
2
  "schema_version": 1,
3
- "_comment": "NV-EmbedCode-7B (MistralBiDirectional + avg-pool) on Hexagon v81. 7B encoder split into 4 chained W8 context parts via llama_embed_sharded. Host-rope (no Gather), PLAIN rope (Mistral theta 10000, factor 1), bidirectional. Encoder S=128, head_dim 128.",
4
- "model": {"name": "nv-embedcode-7b", "family": "embedding", "dsp_arch": "v81", "tokenizer_pre": "default"},
5
- "params": {"hidden": 4096, "vocab": 32000, "n_layers": 32, "head_dim": 128, "max_ctx": 128, "eos_token_id": 2},
 
 
 
 
 
 
 
 
 
 
 
6
  "artifacts": {
7
  "contexts": {
8
- "enc_p0": {"bin": "nvembedcode7b_enc_p0_w8.bin"},
9
- "enc_p1": {"bin": "nvembedcode7b_enc_p1_w8.bin"},
10
- "enc_p2": {"bin": "nvembedcode7b_enc_p2_w8.bin"},
11
- "enc_p3": {"bin": "nvembedcode7b_enc_p3_w8.bin"}
 
 
 
 
 
 
 
 
12
  },
13
  "embed": "nvembedcode7b_embed_f16.bin",
14
  "tokenizer": "tokenizer.json"
15
  },
16
- "plan": {"steps": [{"host": "llama_embed_sharded", "params": {"encoder_parts": "nvembedcode7b_enc_p0_w8,nvembedcode7b_enc_p1_w8,nvembedcode7b_enc_p2_w8,nvembedcode7b_enc_p3_w8", "rope_factor": 1.0, "rope_theta": 10000.0, "neg": -50000}}]}
 
 
 
 
 
 
 
 
 
 
 
 
17
  }
 
1
  {
2
  "schema_version": 1,
3
+ "model": {
4
+ "name": "nv-embedcode-7b",
5
+ "family": "embedding",
6
+ "dsp_arch": "v81",
7
+ "tokenizer_pre": "default"
8
+ },
9
+ "params": {
10
+ "hidden": 4096,
11
+ "vocab": 32000,
12
+ "n_layers": 32,
13
+ "head_dim": 128,
14
+ "max_ctx": 128,
15
+ "eos_token_id": 2
16
+ },
17
  "artifacts": {
18
  "contexts": {
19
+ "enc_p0": {
20
+ "bin": "nvembedcode7b_enc_p0_w8.bin"
21
+ },
22
+ "enc_p1": {
23
+ "bin": "nvembedcode7b_enc_p1_w8.bin"
24
+ },
25
+ "enc_p2": {
26
+ "bin": "nvembedcode7b_enc_p2_w8.bin"
27
+ },
28
+ "enc_p3": {
29
+ "bin": "nvembedcode7b_enc_p3_w8.bin"
30
+ }
31
  },
32
  "embed": "nvembedcode7b_embed_f16.bin",
33
  "tokenizer": "tokenizer.json"
34
  },
35
+ "plan": {
36
+ "steps": [
37
+ {
38
+ "host": "llama_embed_sharded",
39
+ "params": {
40
+ "encoder_parts": "nvembedcode7b_enc_p0_w8,nvembedcode7b_enc_p1_w8,nvembedcode7b_enc_p2_w8,nvembedcode7b_enc_p3_w8",
41
+ "rope_factor": 1.0,
42
+ "rope_theta": 10000.0,
43
+ "neg": -50000
44
+ }
45
+ }
46
+ ]
47
+ }
48
  }