mlboydaisuke commited on
Commit
7c50ac6
·
verified ·
1 Parent(s): add26ee

Chat template: accept the 0.18 content-parts form (string form unchanged); weights, tokenizer and executor metadata byte-identical

Browse files
Files changed (3) hide show
  1. README.md +2 -0
  2. S1-mini_int8.litertlm +1 -1
  3. litertlm_manifest.json +3 -3
README.md CHANGED
@@ -36,6 +36,8 @@ S1-mini takes a raw speech-to-text transcript and rewrites it as clean written t
36
  |---|---|---|
37
  | `S1-mini_int8.litertlm` | int8 dynamic on linears + embedding (`dynamic_wi8_afp32`) | 688 MB |
38
 
 
 
39
  int8 only, and that is a measurement rather than a default. An int4 blockwise-32 build was made and rejected: it drops the comma before "and", drops discourse words like "hmm", and decodes *slower* than int8 at this size, on the same Mac and the same harness. Punctuation fidelity is this model's entire product.
40
 
41
  ## Usage
 
36
  |---|---|---|
37
  | `S1-mini_int8.litertlm` | int8 dynamic on linears + embedding (`dynamic_wi8_afp32`) | 688 MB |
38
 
39
+ 2026-09-21: chat template updated to accept the 0.18 content-parts form (string form unchanged); weights, tokenizer and executor metadata byte-identical.
40
+
41
  int8 only, and that is a measurement rather than a default. An int4 blockwise-32 build was made and rejected: it drops the comma before "and", drops discourse words like "hmm", and decodes *slower* than int8 at this size, on the same Mac and the same harness. Punctuation fidelity is this model's entire product.
42
 
43
  ## Usage
S1-mini_int8.litertlm CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0376f042102cd1409055374d3e823faaaf3a66c5f61c6c744ffe01193d0ae397
3
  size 688157792
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:94828f0da7774e36acaedf36bfbc54ec9d81a3076ea10ed3c5e45f4ccbdb94e5
3
  size 688157792
litertlm_manifest.json CHANGED
@@ -1,7 +1,7 @@
1
  {
2
  "manifest_schema": "0.1.2",
3
  "repo": "mlboydaisuke/S1-mini-LiteRT",
4
- "generated": "2026-09-06",
5
  "generator": "make_manifest.py",
6
  "model": {
7
  "display_name": "S1-mini (LiteRT-LM)",
@@ -35,12 +35,12 @@
35
  "variants": [
36
  {
37
  "file": "S1-mini_int8.litertlm",
38
- "sha256": "0376f042102cd1409055374d3e823faaaf3a66c5f61c6c744ffe01193d0ae397",
39
  "size_bytes": 688157792,
40
  "sections": [
41
  {
42
  "type": "LlmMetadataProto",
43
- "size_bytes": 770
44
  },
45
  {
46
  "type": "HF_Tokenizer_Zlib",
 
1
  {
2
  "manifest_schema": "0.1.2",
3
  "repo": "mlboydaisuke/S1-mini-LiteRT",
4
+ "generated": "2026-09-22",
5
  "generator": "make_manifest.py",
6
  "model": {
7
  "display_name": "S1-mini (LiteRT-LM)",
 
35
  "variants": [
36
  {
37
  "file": "S1-mini_int8.litertlm",
38
+ "sha256": "94828f0da7774e36acaedf36bfbc54ec9d81a3076ea10ed3c5e45f4ccbdb94e5",
39
  "size_bytes": 688157792,
40
  "sections": [
41
  {
42
  "type": "LlmMetadataProto",
43
+ "size_bytes": 1137
44
  },
45
  {
46
  "type": "HF_Tokenizer_Zlib",