Upload Zoder 2.0-1B REFINED (TIES + SLERP Pullback)
Browse files
README.md
CHANGED
|
@@ -1,64 +1,42 @@
|
|
| 1 |
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
tags: [mergekit, ties, slerp, minicpm, zoder, gguf]
|
| 4 |
base_model:
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
-
|
| 10 |
-
|
| 11 |
-
# 🧬 Zoder 2.0-1B (REFINED)
|
| 12 |
-
|
| 13 |
-
Two-phase refinement: TIES ensemble + SLERP pullback toward the proven Zoder 1.0-1B lineage, with auto-tuned interpolation.
|
| 14 |
-
|
| 15 |
-
## Merge Architecture
|
| 16 |
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
├── V2 (w0.8, d0.9) ─┐
|
| 20 |
-
── V1 (w0.2, d0.6) ─┤ TIES → Phase-1
|
| 21 |
-
│
|
| 22 |
-
Zoder1.0-1B ─── SLERP(t=0.2) ─── Phase-1 → ZODER 2.0-1B REFINED
|
| 23 |
-
```
|
| 24 |
-
|
| 25 |
-
## Auto-Tune Sweep Results
|
| 26 |
|
| 27 |
-
|
| 28 |
-
|---|---------|-------------|
|
| 29 |
-
| 0.2 | 80% | 4/4 |
|
| 30 |
-
| 0.3 | 70% | 3/4 |
|
| 31 |
-
| 0.4 | 60% | 4/4 |
|
| 32 |
|
| 33 |
-
|
|
|
|
| 34 |
|
| 35 |
-
|
| 36 |
|
| 37 |
-
|
| 38 |
-
|------|--------|
|
| 39 |
-
| Basic Conversation | ✅ PASS |
|
| 40 |
-
| Logical Reasoning | ✅ PASS |
|
| 41 |
-
| Code Generation | ✅ PASS |
|
| 42 |
-
| Creative Writing | ✅ PASS |
|
| 43 |
-
| Instruction Following | ✅ PASS |
|
| 44 |
-
| Multi-turn Memory | ❌ WEAK |
|
| 45 |
-
| Math Problem | ✅ PASS |
|
| 46 |
-
| Termux Knowledge | ✅ PASS |
|
| 47 |
|
| 48 |
-
|
|
|
|
|
|
|
| 49 |
|
| 50 |
-
##
|
| 51 |
|
| 52 |
-
|
| 53 |
-
|---|---|---|
|
| 54 |
-
| Q2_K | 463.3 MB | 1-2GB RAM (fastest) |
|
| 55 |
-
| Q3_K_M | 555.9 MB | 2-3GB RAM |
|
| 56 |
-
| Q4_K_M | 656.2 MB | Mobile/Termux 4GB+ (sweet spot) |
|
| 57 |
-
| Q5_K_M | 750.4 MB | 8GB+ RAM |
|
| 58 |
-
| Q6_K | 850.5 MB | 12GB+ RAM |
|
| 59 |
-
| Q8_0 | 1100.1 MB | 16GB+ RAM (max quality) |
|
| 60 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
|
| 62 |
-
|
| 63 |
-
- Name: Zoder
|
| 64 |
-
- Model: Zoder2.0-1B
|
|
|
|
| 1 |
---
|
|
|
|
|
|
|
| 2 |
base_model:
|
| 3 |
+
- Nasaawakening/Zoder1.0-1B
|
| 4 |
+
library_name: transformers
|
| 5 |
+
tags:
|
| 6 |
+
- mergekit
|
| 7 |
+
- merge
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
|
| 9 |
+
---
|
| 10 |
+
# zoder2-final
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
|
| 12 |
+
This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
+
## Merge Details
|
| 15 |
+
### Merge Method
|
| 16 |
|
| 17 |
+
This model was merged using the [SLERP](https://en.wikipedia.org/wiki/Slerp) merge method.
|
| 18 |
|
| 19 |
+
### Models Merged
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
|
| 21 |
+
The following models were included in the merge:
|
| 22 |
+
* /kaggle/working/zoder2-phase1
|
| 23 |
+
* [Nasaawakening/Zoder1.0-1B](https://huggingface.co/Nasaawakening/Zoder1.0-1B)
|
| 24 |
|
| 25 |
+
### Configuration
|
| 26 |
|
| 27 |
+
The following YAML configuration was used to produce this model:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
|
| 29 |
+
```yaml
|
| 30 |
+
slices:
|
| 31 |
+
- sources:
|
| 32 |
+
- model: Nasaawakening/Zoder1.0-1B
|
| 33 |
+
layer_range: [0, 24]
|
| 34 |
+
- model: /kaggle/working/zoder2-phase1
|
| 35 |
+
layer_range: [0, 24]
|
| 36 |
+
merge_method: slerp
|
| 37 |
+
base_model: Nasaawakening/Zoder1.0-1B
|
| 38 |
+
parameters:
|
| 39 |
+
t: 0.2
|
| 40 |
+
dtype: float16
|
| 41 |
|
| 42 |
+
```
|
|
|
|
|
|