Bartosz Hanc commited on
Commit ·
097fa68
1
Parent(s): ed5747b
docs: update model card, context lengths, and config specs
Browse files- 0_5b/mlx/config.json +3 -3
- 0_5b/xnnpack/config.json +1 -48
- 1_5b/mlx/config.json +3 -3
- 1_5b/xnnpack/config.json +1 -48
- 3b/mlx/config.json +3 -3
- 3b/xnnpack/config.json +1 -48
- README.md +24 -22
0_5b/mlx/config.json
CHANGED
|
@@ -10,7 +10,7 @@
|
|
| 10 |
"variants": [
|
| 11 |
{
|
| 12 |
"file": "hammer_2_1_0_5b_mlx_int4.pte",
|
| 13 |
-
"precision": "
|
| 14 |
"quantized": true,
|
| 15 |
"default": true,
|
| 16 |
"methods": {
|
|
@@ -40,7 +40,7 @@
|
|
| 40 |
1,
|
| 41 |
151665
|
| 42 |
],
|
| 43 |
-
"dtype": "
|
| 44 |
}
|
| 45 |
]
|
| 46 |
},
|
|
@@ -49,7 +49,7 @@
|
|
| 49 |
151645,
|
| 50 |
151643
|
| 51 |
],
|
| 52 |
-
"get_max_context_len":
|
| 53 |
"get_max_seq_len": 2048,
|
| 54 |
"get_n_layers": 24,
|
| 55 |
"get_vocab_size": 151665,
|
|
|
|
| 10 |
"variants": [
|
| 11 |
{
|
| 12 |
"file": "hammer_2_1_0_5b_mlx_int4.pte",
|
| 13 |
+
"precision": "int4",
|
| 14 |
"quantized": true,
|
| 15 |
"default": true,
|
| 16 |
"methods": {
|
|
|
|
| 40 |
1,
|
| 41 |
151665
|
| 42 |
],
|
| 43 |
+
"dtype": "float32"
|
| 44 |
}
|
| 45 |
]
|
| 46 |
},
|
|
|
|
| 49 |
151645,
|
| 50 |
151643
|
| 51 |
],
|
| 52 |
+
"get_max_context_len": 16384,
|
| 53 |
"get_max_seq_len": 2048,
|
| 54 |
"get_n_layers": 24,
|
| 55 |
"get_vocab_size": 151665,
|
0_5b/xnnpack/config.json
CHANGED
|
@@ -8,53 +8,6 @@
|
|
| 8 |
"backend": "xnnpack",
|
| 9 |
"license": "cc-by-nc-4.0",
|
| 10 |
"variants": [
|
| 11 |
-
{
|
| 12 |
-
"file": "hammer_2_1_0_5b_xnnpack_bf16.pte",
|
| 13 |
-
"precision": "bf16",
|
| 14 |
-
"methods": {
|
| 15 |
-
"enable_dynamic_shape": true,
|
| 16 |
-
"forward": {
|
| 17 |
-
"inputs": [
|
| 18 |
-
{
|
| 19 |
-
"shape": [
|
| 20 |
-
1,
|
| 21 |
-
{
|
| 22 |
-
"min": 1,
|
| 23 |
-
"max": 2047
|
| 24 |
-
}
|
| 25 |
-
],
|
| 26 |
-
"dtype": "int64"
|
| 27 |
-
},
|
| 28 |
-
{
|
| 29 |
-
"shape": [
|
| 30 |
-
1
|
| 31 |
-
],
|
| 32 |
-
"dtype": "int64"
|
| 33 |
-
}
|
| 34 |
-
],
|
| 35 |
-
"outputs": [
|
| 36 |
-
{
|
| 37 |
-
"shape": [
|
| 38 |
-
1,
|
| 39 |
-
151665
|
| 40 |
-
],
|
| 41 |
-
"dtype": "bfloat16"
|
| 42 |
-
}
|
| 43 |
-
]
|
| 44 |
-
},
|
| 45 |
-
"get_bos_id": 151643,
|
| 46 |
-
"get_eos_ids": [
|
| 47 |
-
151645,
|
| 48 |
-
151643
|
| 49 |
-
],
|
| 50 |
-
"get_max_context_len": 2048,
|
| 51 |
-
"get_max_seq_len": 2048,
|
| 52 |
-
"get_n_layers": 24,
|
| 53 |
-
"get_vocab_size": 151665,
|
| 54 |
-
"use_kv_cache": true,
|
| 55 |
-
"use_sdpa_with_kv_cache": true
|
| 56 |
-
}
|
| 57 |
-
},
|
| 58 |
{
|
| 59 |
"file": "hammer_2_1_0_5b_xnnpack_8da4w.pte",
|
| 60 |
"precision": "8da4w",
|
|
@@ -94,7 +47,7 @@
|
|
| 94 |
151645,
|
| 95 |
151643
|
| 96 |
],
|
| 97 |
-
"get_max_context_len":
|
| 98 |
"get_max_seq_len": 2048,
|
| 99 |
"get_n_layers": 24,
|
| 100 |
"get_vocab_size": 151665,
|
|
|
|
| 8 |
"backend": "xnnpack",
|
| 9 |
"license": "cc-by-nc-4.0",
|
| 10 |
"variants": [
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
{
|
| 12 |
"file": "hammer_2_1_0_5b_xnnpack_8da4w.pte",
|
| 13 |
"precision": "8da4w",
|
|
|
|
| 47 |
151645,
|
| 48 |
151643
|
| 49 |
],
|
| 50 |
+
"get_max_context_len": 16384,
|
| 51 |
"get_max_seq_len": 2048,
|
| 52 |
"get_n_layers": 24,
|
| 53 |
"get_vocab_size": 151665,
|
1_5b/mlx/config.json
CHANGED
|
@@ -10,7 +10,7 @@
|
|
| 10 |
"variants": [
|
| 11 |
{
|
| 12 |
"file": "hammer_2_1_1_5b_mlx_int4.pte",
|
| 13 |
-
"precision": "
|
| 14 |
"quantized": true,
|
| 15 |
"default": true,
|
| 16 |
"methods": {
|
|
@@ -40,7 +40,7 @@
|
|
| 40 |
1,
|
| 41 |
151665
|
| 42 |
],
|
| 43 |
-
"dtype": "
|
| 44 |
}
|
| 45 |
]
|
| 46 |
},
|
|
@@ -49,7 +49,7 @@
|
|
| 49 |
151645,
|
| 50 |
151643
|
| 51 |
],
|
| 52 |
-
"get_max_context_len":
|
| 53 |
"get_max_seq_len": 2048,
|
| 54 |
"get_n_layers": 28,
|
| 55 |
"get_vocab_size": 151665,
|
|
|
|
| 10 |
"variants": [
|
| 11 |
{
|
| 12 |
"file": "hammer_2_1_1_5b_mlx_int4.pte",
|
| 13 |
+
"precision": "int4",
|
| 14 |
"quantized": true,
|
| 15 |
"default": true,
|
| 16 |
"methods": {
|
|
|
|
| 40 |
1,
|
| 41 |
151665
|
| 42 |
],
|
| 43 |
+
"dtype": "float32"
|
| 44 |
}
|
| 45 |
]
|
| 46 |
},
|
|
|
|
| 49 |
151645,
|
| 50 |
151643
|
| 51 |
],
|
| 52 |
+
"get_max_context_len": 16384,
|
| 53 |
"get_max_seq_len": 2048,
|
| 54 |
"get_n_layers": 28,
|
| 55 |
"get_vocab_size": 151665,
|
1_5b/xnnpack/config.json
CHANGED
|
@@ -8,53 +8,6 @@
|
|
| 8 |
"backend": "xnnpack",
|
| 9 |
"license": "cc-by-nc-4.0",
|
| 10 |
"variants": [
|
| 11 |
-
{
|
| 12 |
-
"file": "hammer_2_1_1_5b_xnnpack_bf16.pte",
|
| 13 |
-
"precision": "bf16",
|
| 14 |
-
"methods": {
|
| 15 |
-
"enable_dynamic_shape": true,
|
| 16 |
-
"forward": {
|
| 17 |
-
"inputs": [
|
| 18 |
-
{
|
| 19 |
-
"shape": [
|
| 20 |
-
1,
|
| 21 |
-
{
|
| 22 |
-
"min": 1,
|
| 23 |
-
"max": 2047
|
| 24 |
-
}
|
| 25 |
-
],
|
| 26 |
-
"dtype": "int64"
|
| 27 |
-
},
|
| 28 |
-
{
|
| 29 |
-
"shape": [
|
| 30 |
-
1
|
| 31 |
-
],
|
| 32 |
-
"dtype": "int64"
|
| 33 |
-
}
|
| 34 |
-
],
|
| 35 |
-
"outputs": [
|
| 36 |
-
{
|
| 37 |
-
"shape": [
|
| 38 |
-
1,
|
| 39 |
-
151665
|
| 40 |
-
],
|
| 41 |
-
"dtype": "bfloat16"
|
| 42 |
-
}
|
| 43 |
-
]
|
| 44 |
-
},
|
| 45 |
-
"get_bos_id": 151643,
|
| 46 |
-
"get_eos_ids": [
|
| 47 |
-
151645,
|
| 48 |
-
151643
|
| 49 |
-
],
|
| 50 |
-
"get_max_context_len": 2048,
|
| 51 |
-
"get_max_seq_len": 2048,
|
| 52 |
-
"get_n_layers": 28,
|
| 53 |
-
"get_vocab_size": 151665,
|
| 54 |
-
"use_kv_cache": true,
|
| 55 |
-
"use_sdpa_with_kv_cache": true
|
| 56 |
-
}
|
| 57 |
-
},
|
| 58 |
{
|
| 59 |
"file": "hammer_2_1_1_5b_xnnpack_8da4w.pte",
|
| 60 |
"precision": "8da4w",
|
|
@@ -94,7 +47,7 @@
|
|
| 94 |
151645,
|
| 95 |
151643
|
| 96 |
],
|
| 97 |
-
"get_max_context_len":
|
| 98 |
"get_max_seq_len": 2048,
|
| 99 |
"get_n_layers": 28,
|
| 100 |
"get_vocab_size": 151665,
|
|
|
|
| 8 |
"backend": "xnnpack",
|
| 9 |
"license": "cc-by-nc-4.0",
|
| 10 |
"variants": [
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
{
|
| 12 |
"file": "hammer_2_1_1_5b_xnnpack_8da4w.pte",
|
| 13 |
"precision": "8da4w",
|
|
|
|
| 47 |
151645,
|
| 48 |
151643
|
| 49 |
],
|
| 50 |
+
"get_max_context_len": 16384,
|
| 51 |
"get_max_seq_len": 2048,
|
| 52 |
"get_n_layers": 28,
|
| 53 |
"get_vocab_size": 151665,
|
3b/mlx/config.json
CHANGED
|
@@ -10,7 +10,7 @@
|
|
| 10 |
"variants": [
|
| 11 |
{
|
| 12 |
"file": "hammer_2_1_3b_mlx_int4.pte",
|
| 13 |
-
"precision": "
|
| 14 |
"quantized": true,
|
| 15 |
"default": true,
|
| 16 |
"methods": {
|
|
@@ -40,7 +40,7 @@
|
|
| 40 |
1,
|
| 41 |
151665
|
| 42 |
],
|
| 43 |
-
"dtype": "
|
| 44 |
}
|
| 45 |
]
|
| 46 |
},
|
|
@@ -49,7 +49,7 @@
|
|
| 49 |
151645,
|
| 50 |
151643
|
| 51 |
],
|
| 52 |
-
"get_max_context_len":
|
| 53 |
"get_max_seq_len": 2048,
|
| 54 |
"get_n_layers": 36,
|
| 55 |
"get_vocab_size": 151665,
|
|
|
|
| 10 |
"variants": [
|
| 11 |
{
|
| 12 |
"file": "hammer_2_1_3b_mlx_int4.pte",
|
| 13 |
+
"precision": "int4",
|
| 14 |
"quantized": true,
|
| 15 |
"default": true,
|
| 16 |
"methods": {
|
|
|
|
| 40 |
1,
|
| 41 |
151665
|
| 42 |
],
|
| 43 |
+
"dtype": "float32"
|
| 44 |
}
|
| 45 |
]
|
| 46 |
},
|
|
|
|
| 49 |
151645,
|
| 50 |
151643
|
| 51 |
],
|
| 52 |
+
"get_max_context_len": 8192,
|
| 53 |
"get_max_seq_len": 2048,
|
| 54 |
"get_n_layers": 36,
|
| 55 |
"get_vocab_size": 151665,
|
3b/xnnpack/config.json
CHANGED
|
@@ -8,53 +8,6 @@
|
|
| 8 |
"backend": "xnnpack",
|
| 9 |
"license": "cc-by-nc-4.0",
|
| 10 |
"variants": [
|
| 11 |
-
{
|
| 12 |
-
"file": "hammer_2_1_3b_xnnpack_bf16.pte",
|
| 13 |
-
"precision": "bf16",
|
| 14 |
-
"methods": {
|
| 15 |
-
"enable_dynamic_shape": true,
|
| 16 |
-
"forward": {
|
| 17 |
-
"inputs": [
|
| 18 |
-
{
|
| 19 |
-
"shape": [
|
| 20 |
-
1,
|
| 21 |
-
{
|
| 22 |
-
"min": 1,
|
| 23 |
-
"max": 2047
|
| 24 |
-
}
|
| 25 |
-
],
|
| 26 |
-
"dtype": "int64"
|
| 27 |
-
},
|
| 28 |
-
{
|
| 29 |
-
"shape": [
|
| 30 |
-
1
|
| 31 |
-
],
|
| 32 |
-
"dtype": "int64"
|
| 33 |
-
}
|
| 34 |
-
],
|
| 35 |
-
"outputs": [
|
| 36 |
-
{
|
| 37 |
-
"shape": [
|
| 38 |
-
1,
|
| 39 |
-
151665
|
| 40 |
-
],
|
| 41 |
-
"dtype": "bfloat16"
|
| 42 |
-
}
|
| 43 |
-
]
|
| 44 |
-
},
|
| 45 |
-
"get_bos_id": 151643,
|
| 46 |
-
"get_eos_ids": [
|
| 47 |
-
151645,
|
| 48 |
-
151643
|
| 49 |
-
],
|
| 50 |
-
"get_max_context_len": 2048,
|
| 51 |
-
"get_max_seq_len": 2048,
|
| 52 |
-
"get_n_layers": 36,
|
| 53 |
-
"get_vocab_size": 151665,
|
| 54 |
-
"use_kv_cache": true,
|
| 55 |
-
"use_sdpa_with_kv_cache": true
|
| 56 |
-
}
|
| 57 |
-
},
|
| 58 |
{
|
| 59 |
"file": "hammer_2_1_3b_xnnpack_8da4w.pte",
|
| 60 |
"precision": "8da4w",
|
|
@@ -94,7 +47,7 @@
|
|
| 94 |
151645,
|
| 95 |
151643
|
| 96 |
],
|
| 97 |
-
"get_max_context_len":
|
| 98 |
"get_max_seq_len": 2048,
|
| 99 |
"get_n_layers": 36,
|
| 100 |
"get_vocab_size": 151665,
|
|
|
|
| 8 |
"backend": "xnnpack",
|
| 9 |
"license": "cc-by-nc-4.0",
|
| 10 |
"variants": [
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
{
|
| 12 |
"file": "hammer_2_1_3b_xnnpack_8da4w.pte",
|
| 13 |
"precision": "8da4w",
|
|
|
|
| 47 |
151645,
|
| 48 |
151643
|
| 49 |
],
|
| 50 |
+
"get_max_context_len": 8192,
|
| 51 |
"get_max_seq_len": 2048,
|
| 52 |
"get_n_layers": 36,
|
| 53 |
"get_vocab_size": 151665,
|
README.md
CHANGED
|
@@ -14,33 +14,35 @@ library as ExecuTorch `.pte` programs, ready to run on device.
|
|
| 14 |
|
| 15 |
## Variants
|
| 16 |
|
| 17 |
-
| Path | Backend | Precision |
|
| 18 |
-
| --- | --- | --- |
|
| 19 |
-
| `0_5b/mlx/hammer_2_1_0_5b_mlx_int4.pte` | mlx |
|
| 20 |
-
| `0_5b/xnnpack/hammer_2_1_0_5b_xnnpack_8da4w.pte` | xnnpack | 8da4w |
|
| 21 |
-
| `1_5b/mlx/hammer_2_1_1_5b_mlx_int4.pte` | mlx |
|
| 22 |
-
| `1_5b/xnnpack/hammer_2_1_1_5b_xnnpack_8da4w.pte` | xnnpack | 8da4w |
|
| 23 |
-
| `3b/mlx/hammer_2_1_3b_mlx_int4.pte` | mlx |
|
| 24 |
-
| `3b/xnnpack/hammer_2_1_3b_xnnpack_8da4w.pte` | xnnpack | 8da4w |
|
|
|
|
|
|
|
| 25 |
|
| 26 |
## Repository structure
|
| 27 |
|
| 28 |
```
|
| 29 |
-
0_5b/mlx/config.json 1.3
|
| 30 |
-
0_5b/mlx/hammer_2_1_0_5b_mlx_int4.pte 510
|
| 31 |
-
0_5b/xnnpack/config.json
|
| 32 |
-
0_5b/xnnpack/hammer_2_1_0_5b_xnnpack_8da4w.pte
|
| 33 |
-
1_5b/mlx/config.json 1.3
|
| 34 |
-
1_5b/mlx/hammer_2_1_1_5b_mlx_int4.pte 1.2
|
| 35 |
-
1_5b/xnnpack/config.json
|
| 36 |
-
1_5b/xnnpack/hammer_2_1_1_5b_xnnpack_8da4w.pte 1.1
|
| 37 |
-
3b/mlx/config.json 1.3
|
| 38 |
-
3b/mlx/hammer_2_1_3b_mlx_int4.pte 2.1
|
| 39 |
-
3b/xnnpack/config.json
|
| 40 |
-
3b/xnnpack/hammer_2_1_3b_xnnpack_8da4w.pte 1.9
|
| 41 |
config.json 32 B
|
| 42 |
-
tokenizer.json 10.9
|
| 43 |
-
tokenizer_config.json 7.0
|
| 44 |
```
|
| 45 |
|
| 46 |
## Compatibility
|
|
|
|
| 14 |
|
| 15 |
## Variants
|
| 16 |
|
| 17 |
+
| Path | Backend | Precision | Context length |
|
| 18 |
+
| --- | --- | --- | --- |
|
| 19 |
+
| `0_5b/mlx/hammer_2_1_0_5b_mlx_int4.pte` | mlx | int4 | 16,384 |
|
| 20 |
+
| `0_5b/xnnpack/hammer_2_1_0_5b_xnnpack_8da4w.pte` | xnnpack | 8da4w | 16,384 |
|
| 21 |
+
| `1_5b/mlx/hammer_2_1_1_5b_mlx_int4.pte` | mlx | int4 | 16,384 |
|
| 22 |
+
| `1_5b/xnnpack/hammer_2_1_1_5b_xnnpack_8da4w.pte` | xnnpack | 8da4w | 16,384 |
|
| 23 |
+
| `3b/mlx/hammer_2_1_3b_mlx_int4.pte` | mlx | int4 | 8,192 |
|
| 24 |
+
| `3b/xnnpack/hammer_2_1_3b_xnnpack_8da4w.pte` | xnnpack | 8da4w | 8,192 |
|
| 25 |
+
|
| 26 |
+
|
| 27 |
|
| 28 |
## Repository structure
|
| 29 |
|
| 30 |
```
|
| 31 |
+
0_5b/mlx/config.json 1.3 KiB
|
| 32 |
+
0_5b/mlx/hammer_2_1_0_5b_mlx_int4.pte 510 MiB
|
| 33 |
+
0_5b/xnnpack/config.json 1.3 KiB
|
| 34 |
+
0_5b/xnnpack/hammer_2_1_0_5b_xnnpack_8da4w.pte 412 MiB
|
| 35 |
+
1_5b/mlx/config.json 1.3 KiB
|
| 36 |
+
1_5b/mlx/hammer_2_1_1_5b_mlx_int4.pte 1.2 GiB
|
| 37 |
+
1_5b/xnnpack/config.json 1.3 KiB
|
| 38 |
+
1_5b/xnnpack/hammer_2_1_1_5b_xnnpack_8da4w.pte 1.1 GiB
|
| 39 |
+
3b/mlx/config.json 1.3 KiB
|
| 40 |
+
3b/mlx/hammer_2_1_3b_mlx_int4.pte 2.1 GiB
|
| 41 |
+
3b/xnnpack/config.json 1.3 KiB
|
| 42 |
+
3b/xnnpack/hammer_2_1_3b_xnnpack_8da4w.pte 1.9 GiB
|
| 43 |
config.json 32 B
|
| 44 |
+
tokenizer.json 10.9 MiB
|
| 45 |
+
tokenizer_config.json 7.0 KiB
|
| 46 |
```
|
| 47 |
|
| 48 |
## Compatibility
|