mlboydaisuke commited on
Commit
013a833
·
verified ·
1 Parent(s): 02e8a86

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

Browse files
MiniCPM5-2B_int4.litertlm CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9858563beafbc6d5e0d25fcee3827541515296a9302ed3d088b16a58d4fbe7b8
3
  size 1553670064
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6caaa7622f8d382a0d7bc18762c58ff37b8fe076d04916ded2f75d5ffd00b7c9
3
  size 1553670064
MiniCPM5-2B_int8.litertlm CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:58da116a6eb563185eb5e8d0b1f1e1116b632a516cadc0285b6da35b9f886a86
3
  size 2603278256
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b0e972ab52fbab3ecfaeaffed00e4a47415620b71f3f53a80249b0dda19142a2
3
  size 2603278256
README.md CHANGED
@@ -26,6 +26,8 @@ MiniCPM5-2B is OpenBMB's 2.5B-parameter dense **hybrid-reasoning** model (42 lay
26
  | `MiniCPM5-2B_int4.litertlm` | int4 blockwise-32 + OCTAV on linears, int8 embedding | 1.55 GB |
27
  | `MiniCPM5-2B_int8.litertlm` | int8 dynamic on linears + embedding; **fp32 activations declared** (see notes) | 2.60 GB |
28
 
 
 
29
  The **int4 file is the phone file** (smaller, fastest GPU decode on every platform measured) — best used for direct answers or short reasoning; see the thinking-mode note below. **int8 is the file for reasoning that has to complete**: its thinking chains are ~3–4× shorter than int4's on the same questions and terminate where int4 runs into the token budget. int8's main weight section is 2.33 GB, above the single-section mmap ceiling of default-entitlement iOS apps, so it is a desktop / Android build.
30
 
31
  ## Correctness
 
26
  | `MiniCPM5-2B_int4.litertlm` | int4 blockwise-32 + OCTAV on linears, int8 embedding | 1.55 GB |
27
  | `MiniCPM5-2B_int8.litertlm` | int8 dynamic on linears + embedding; **fp32 activations declared** (see notes) | 2.60 GB |
28
 
29
+ 2026-09-21: chat template updated to accept the 0.18 content-parts form (string form unchanged); weights, tokenizer and executor metadata byte-identical.
30
+
31
  The **int4 file is the phone file** (smaller, fastest GPU decode on every platform measured) — best used for direct answers or short reasoning; see the thinking-mode note below. **int8 is the file for reasoning that has to complete**: its thinking chains are ~3–4× shorter than int4's on the same questions and terminate where int4 runs into the token budget. int8's main weight section is 2.33 GB, above the single-section mmap ceiling of default-entitlement iOS apps, so it is a desktop / Android build.
32
 
33
  ## Correctness
litertlm_manifest.json CHANGED
@@ -1,7 +1,7 @@
1
  {
2
  "manifest_schema": "0.1.2",
3
  "repo": "mlboydaisuke/MiniCPM5-2B-LiteRT",
4
- "generated": "2026-09-08",
5
  "generator": "make_manifest.py",
6
  "model": {
7
  "display_name": "MiniCPM5-2B",
@@ -32,12 +32,12 @@
32
  "variants": [
33
  {
34
  "file": "MiniCPM5-2B_int4.litertlm",
35
- "sha256": "9858563beafbc6d5e0d25fcee3827541515296a9302ed3d088b16a58d4fbe7b8",
36
  "size_bytes": 1553670064,
37
  "sections": [
38
  {
39
  "type": "LlmMetadataProto",
40
- "size_bytes": 9346
41
  },
42
  {
43
  "type": "HF_Tokenizer_Zlib",
@@ -162,12 +162,12 @@
162
  },
163
  {
164
  "file": "MiniCPM5-2B_int8.litertlm",
165
- "sha256": "58da116a6eb563185eb5e8d0b1f1e1116b632a516cadc0285b6da35b9f886a86",
166
  "size_bytes": 2603278256,
167
  "sections": [
168
  {
169
  "type": "LlmMetadataProto",
170
- "size_bytes": 9346
171
  },
172
  {
173
  "type": "HF_Tokenizer_Zlib",
 
1
  {
2
  "manifest_schema": "0.1.2",
3
  "repo": "mlboydaisuke/MiniCPM5-2B-LiteRT",
4
+ "generated": "2026-09-22",
5
  "generator": "make_manifest.py",
6
  "model": {
7
  "display_name": "MiniCPM5-2B",
 
32
  "variants": [
33
  {
34
  "file": "MiniCPM5-2B_int4.litertlm",
35
+ "sha256": "6caaa7622f8d382a0d7bc18762c58ff37b8fe076d04916ded2f75d5ffd00b7c9",
36
  "size_bytes": 1553670064,
37
  "sections": [
38
  {
39
  "type": "LlmMetadataProto",
40
+ "size_bytes": 9857
41
  },
42
  {
43
  "type": "HF_Tokenizer_Zlib",
 
162
  },
163
  {
164
  "file": "MiniCPM5-2B_int8.litertlm",
165
+ "sha256": "b0e972ab52fbab3ecfaeaffed00e4a47415620b71f3f53a80249b0dda19142a2",
166
  "size_bytes": 2603278256,
167
  "sections": [
168
  {
169
  "type": "LlmMetadataProto",
170
+ "size_bytes": 9857
171
  },
172
  {
173
  "type": "HF_Tokenizer_Zlib",