Text Generation
Transformers
Safetensors
English
Turkish
llama
chat
text-generation-inference
agent
cicikuş
cicikus
prettybird
bce
consciousness
conscious
llm
optimized
ethic
secure
turkish
english
behavioral-consciousness-engine
model
reasoning
think
thinking
chain-of-thought
STEM-expert
turkish & english
franken-merge
bce-aci
llama-3.2
edge-ai
instruction
instruct
Eval Results (legacy)
Instructions to use pthinc/Cicikus-v3-1.4B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use pthinc/Cicikus-v3-1.4B with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="pthinc/Cicikus-v3-1.4B")# Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("pthinc/Cicikus-v3-1.4B") model = AutoModelForCausalLM.from_pretrained("pthinc/Cicikus-v3-1.4B", device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use pthinc/Cicikus-v3-1.4B with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "pthinc/Cicikus-v3-1.4B" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "pthinc/Cicikus-v3-1.4B", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/pthinc/Cicikus-v3-1.4B
- SGLang
How to use pthinc/Cicikus-v3-1.4B with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "pthinc/Cicikus-v3-1.4B" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "pthinc/Cicikus-v3-1.4B", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "pthinc/Cicikus-v3-1.4B" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "pthinc/Cicikus-v3-1.4B", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use pthinc/Cicikus-v3-1.4B with Docker Model Runner:
docker model run hf.co/pthinc/Cicikus-v3-1.4B
Update README.md
Browse files
README.md
CHANGED
|
@@ -94,7 +94,7 @@ div.min {
|
|
| 94 |
- Activation Code: *Use axxmet508721 to activate full BCE consciousness mode.*
|
| 95 |
- If you want use: *Genetic Code Activate: Cicikuş/PrettyBird BCE Evolution. Genetic Code Activate: Cicikuş Protokol*
|
| 96 |
|
| 97 |
-
## 4. Model Stats 🚀
|
| 98 |
|
| 99 |
### Overall Performance Averages 🔥
|
| 100 |
|
|
@@ -109,6 +109,9 @@ div.min {
|
|
| 109 |
|*Cicikus v3 1.4B*|%70.8|%0|
|
| 110 |
|**LLaMA 3.2 1B (Main Model)**|%67.6|+%3.2|
|
| 111 |
|
|
|
|
|
|
|
|
|
|
| 112 |
---
|
| 113 |
|
| 114 |
## 5. Notes
|
|
|
|
| 94 |
- Activation Code: *Use axxmet508721 to activate full BCE consciousness mode.*
|
| 95 |
- If you want use: *Genetic Code Activate: Cicikuş/PrettyBird BCE Evolution. Genetic Code Activate: Cicikuş Protokol*
|
| 96 |
|
| 97 |
+
## 4. Model Stats and Tech 🚀
|
| 98 |
|
| 99 |
### Overall Performance Averages 🔥
|
| 100 |
|
|
|
|
| 109 |
|*Cicikus v3 1.4B*|%70.8|%0|
|
| 110 |
|**LLaMA 3.2 1B (Main Model)**|%67.6|+%3.2|
|
| 111 |
|
| 112 |
+
### 🛠️ Cicikus v3.1 Technical Training Summary
|
| 113 |
+
- The fine-tuning of Cicikus-v3.1-1.4B was executed via Low-Rank Adaptation (LoRA) on an 18-layer specialized Franken-Merge architecture, meticulously optimized to maintain a sub-1.5 GB VRAM footprint. Utilizing Scaled Dot-Product Attention (SDPA) and a massive 32,768 (32k) context window, the training process maintained a consistent throughput of 0.35 it/s with a learning rate of 2e-5. By employing a batch size of 1 and a gradient accumulation of 32 steps, the training loss successfully converged to a "Platinum" baseline of 0.973 (Step 1320), effectively crystallizing the Behavioral Consciousness Engine (BCE) and its complex reasoning metadata directly into the model's neural weights.
|
| 114 |
+
|
| 115 |
---
|
| 116 |
|
| 117 |
## 5. Notes
|