Publish Mercan 0.8B SFT Q4 model
Browse files- .gitattributes +1 -0
- README.md +59 -0
- model.mercan +3 -0
- model.mercan.json +1 -0
- model.mercan.sha256 +1 -0
.gitattributes
CHANGED
|
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
+
model.mercan filter=lfs diff=lfs merge=lfs -text
|
README.md
ADDED
|
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- tr
|
| 4 |
+
library_name: mercan
|
| 5 |
+
pipeline_tag: text-generation
|
| 6 |
+
tags:
|
| 7 |
+
- mercan
|
| 8 |
+
- nedolm
|
| 9 |
+
- gguf
|
| 10 |
+
- turkish
|
| 11 |
+
- text-generation
|
| 12 |
+
license: apache-2.0
|
| 13 |
+
---
|
| 14 |
+
|
| 15 |
+
# Mercan 0.8B SFT
|
| 16 |
+
|
| 17 |
+
Mercan 0.8B SFT is a Turkish assistant model distributed in the self-contained Mercan v1 deployment format.
|
| 18 |
+
|
| 19 |
+
## Run
|
| 20 |
+
|
| 21 |
+
Install the Mercan CLI, then run this repository directly:
|
| 22 |
+
|
| 23 |
+
```bash
|
| 24 |
+
mercan run MercanAI/Mercan-0.8B-SFT
|
| 25 |
+
```
|
| 26 |
+
|
| 27 |
+
The CLI downloads `model.mercan` once into the local Mercan cache and then runs it locally.
|
| 28 |
+
|
| 29 |
+
You can also download the artifact yourself:
|
| 30 |
+
|
| 31 |
+
```bash
|
| 32 |
+
mercan run model.mercan
|
| 33 |
+
```
|
| 34 |
+
|
| 35 |
+
## Artifact
|
| 36 |
+
|
| 37 |
+
- `model.mercan`: Q4_K_M deployment model
|
| 38 |
+
- Mercan format v1
|
| 39 |
+
- physical tensor container: GGUF v3
|
| 40 |
+
- exact NDSRF004 tokenizer asset embedded in the model
|
| 41 |
+
- context length: 4096
|
| 42 |
+
- 24 transformer blocks
|
| 43 |
+
- hidden size: 1536
|
| 44 |
+
- 12 attention heads / 4 KV heads
|
| 45 |
+
- MorphFFN in the first 18 blocks
|
| 46 |
+
|
| 47 |
+
The `.mercan` file contains model weights, architecture metadata, tokenizer metadata/data and the token-role routing table. It does not contain executable model-bundled code.
|
| 48 |
+
|
| 49 |
+
## Runtime
|
| 50 |
+
|
| 51 |
+
The reference runtime and CLI are published in `Ahmet2001/mercanApp-test1`. `libmercan` uses a stable C ABI over the native inference backend.
|
| 52 |
+
|
| 53 |
+
## Chat template
|
| 54 |
+
|
| 55 |
+
```text
|
| 56 |
+
<|im_start|>{role}\n{content}<|im_end|>\n
|
| 57 |
+
```
|
| 58 |
+
|
| 59 |
+
EOS is appended once at the end of a training conversation.
|
model.mercan
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a0c717a2f47a24bb064070f625126442eb0cb14123d018ac655522859dd59844
|
| 3 |
+
size 506249792
|
model.mercan.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"format":"mercan","format_version":1,"quantization":"Q4_K_M","source":"step_00023334.pt"}
|
model.mercan.sha256
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
a0c717a2f47a24bb064070f625126442eb0cb14123d018ac655522859dd59844 /arf/scratch/egitimg16/nedolm_0p8b/mercan_export/model.mercan
|