6e11252276cc7b46fa2ed78b49a03d478c69276ee32460ed1718e0c8643554aa

README.md

@@ -0,0 +1,39 @@
+---
+base_model: Goekdeniz-Guelmez/j.o.s.i.e.v4o-7b-orpo-stage1-v1
+language:
+- en
+license: apache-2.0
+tags:
+- text-generation-inference
+- transformers
+- unsloth
+- qwen2
+- trl
+- orpo
+- KANama
+---
+
+# Goekdeniz-Guelmez/KANama-fineweb-v2-test1
+
+The Model [Goekdeniz-Guelmez/KANama-fineweb-v2-test1](https://huggingface.co/Goekdeniz-Guelmez/KANama-fineweb-v2-test1) was created using KANama.
+
+## Use with KANama
+
+```bash
+pip install KANama, transformers
+```
+
+```python
+from model.handler import from_pretrained, quick_inference
+from transformers import AutoTokenizer
+
+tokenizer = AutoTokenizer.from_pretrained("Doctor-Shotgun/TinyLlama-1.1B-32k")
+model = from_pretrained("path/to/model/folder")
+
+prompt="hello"
+
+input_tokens = tokenizer(prompt, return_tensors="pt").input_ids.to(device)
+
+generated_tokens, generated_text = quick_inference(model, input_tokens, max_new_tokens=50, tokenizer=tokenizer)
+print(generated_text)
+```

config.json

@@ -0,0 +1,23 @@
+{
+    "vocab_size": 152064,
+    "pad_id": 151645,
+    "eos_id": -1,
+    "dim": 256,
+    "n_layers": 18,
+    "n_heads": 12,
+    "n_kv_heads": 6,
+    "use_kan": true,
+    "train_softmax_temp": true,
+    "use_softmax_temp_proj": true,
+    "softmax_bias": false,
+    "multiple_of": 256,
+    "ffn_dim_multiplier": null,
+    "rms_norm_eps": 1e-05,
+    "rope_theta": 500000,
+    "use_scaled_rope": false,
+    "max_batch_size": 100,
+    "max_seq_len": 128,
+    "num_experts": 14,
+    "num_experts_per_tok": 4,
+    "model_type": "KANaMoEv2"
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8506e7111b80c6d8635951a02eab0f4e1a8e4e5772da83846579e97b16f61bf
+size 7031673