Instructions to use nileshc90/apex-phi-14b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use nileshc90/apex-phi-14b with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="nileshc90/apex-phi-14b")

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("nileshc90/apex-phi-14b", device_map="auto")

MLX

How to use nileshc90/apex-phi-14b with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm
# if on a CUDA device, also pip install mlx[cuda]

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("nileshc90/apex-phi-14b")

prompt = "Once upon a time in"
text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

vLLM

How to use nileshc90/apex-phi-14b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "nileshc90/apex-phi-14b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "nileshc90/apex-phi-14b",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/nileshc90/apex-phi-14b

SGLang

How to use nileshc90/apex-phi-14b 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 "nileshc90/apex-phi-14b" \
    --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": "nileshc90/apex-phi-14b",
		"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 "nileshc90/apex-phi-14b" \
        --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": "nileshc90/apex-phi-14b",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

MLX LM

How to use nileshc90/apex-phi-14b with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Generate some text
mlx_lm.generate --model "nileshc90/apex-phi-14b" --prompt "Once upon a time"

Docker Model Runner
How to use nileshc90/apex-phi-14b with Docker Model Runner:
```
docker model run hf.co/nileshc90/apex-phi-14b
```
Atomic Chat

⚡ ApexPhi-14B: Dual-Edition Workstation Suite (Coder & Think)

Microsoft Phi-4 (14.7B) Architecture | Mathematical Invariants & Formal Proofs | Optimized for 36 GB Unified RAM

🏛️ Executive Overview

ApexPhi-14B delivers a specialized dual-edition intelligence architecture built on Microsoft Research's Phi-4 (14.7B), optimized for developer workstations and Apple Silicon Macs (M4 Max / M2 Max):

⚡ ApexPhi-Coder-14B:
- Specialized in high-precision algorithmic coding, AST-verified syntax, and type safety.
- Direct, deterministic execution with zero conversational fluff.
- Zero-Assumption Invariant: strict schema and boundary verification.
- Sampling: temperature 0.2, top_p 0.95, min_p 0.04.
🧠 ApexPhi-Think-14B:
- Specialized in deep mathematical theorem proving, combinatorial logic, and state-machine transitions.
- Employs formal step-by-step invariant derivations inside <think> ... </think> before stating proofs or code.
- Formal verification guardrails: mathematically audits parity, induction steps, and boundary states.
- Sampling: temperature 0.6, top_p 0.95, min_p 0.05.

Both editions run within ~10.4 GB RAM (Q5_K_M) or ~8.9 GB RAM (Q4_K_M), leaving over 23 GB of free host RAM on a 36 GB workstation for IDEs, Docker, and developer tools with 0.00 MB swap churn.

🧠 Why ApexPhi-14B as an Algorithmic Backbone?

Unmatched Mathematical & Algorithmic Density:
- MATH-500: 84.8% (outperforms many 70B models).
- GPQA Diamond: 56.1% graduate-level scientific and algorithmic reasoning.
- Trained on high-density synthetic textbook reasoning data with multi-turn verification.
Zero Non-Western Georestrictions:
- Developed in the United States by Microsoft Research under the permissive MIT License.
- Fully open weights for commercial and enterprise deployment.
Dynamic Min-P Sampling Cutoff:
- Dynamically prunes low-probability hallucinated tail tokens relative to the top token's confidence, ensuring high logical coherence.

💻 36 GB Unified Memory Allocation

+--------------------------------------------------------------------------------+
|                         Total Unified System RAM: 36 GB                        |
+---------------------------------------------------+----------------------------+
|             Active AI Runtime: ~11.0 GB           |    Host Headroom: ~25.0 GB |
|  +---------------------------------------------+  |  +----------------------+  |
|  | ApexPhi-14B (Coder / Think)                 |  |  | macOS Kernel (3.5 GB)|  |
|  | ~10.4 GB (0.0 MB Swap Churn)                |  |  +----------------------+  |
|  +---------------------------------------------+  |  | VS Code / IDE (2.5GB)|  |
|  | Q8_0 KV Cache Buffer (~0.6 GB)              |  |  +----------------------+  |
|  +---------------------------------------------+  |  | Docker & Web (6.0 GB)|  |
|                                                   |  +----------------------+  |
|                                                   |  | Build Cache (13.0 GB)|  |
+---------------------------------------------------+----------------------------+

🚀 1-Click Run Commands

A. Run with Ollama

# 1. Run the High-Precision Algorithmic Coder Edition:
ollama run apex-phi-coder-14b

# 2. Run the Deep Mathematical Proof / Think Edition:
ollama run apex-phi-think-14b

B. High-Throughput Apple Silicon Metal GPU (MLX)

# Serves local OpenAI-compatible endpoint at http://127.0.0.1:8082/v1
bash scripts/serve_mlx_phi4.sh 8082

🛠️ Continue.dev & VS Code Tab Autocomplete Configuration

Add to your ~/.continue/config.json:

{
  "models": [
    {
      "title": "ApexPhi-Coder-14B (High-Precision Algorithmic Coder)",
      "provider": "ollama",
      "model": "apex-phi-coder-14b",
      "apiBase": "http://localhost:11434"
    },
    {
      "title": "ApexPhi-Think-14B (Mathematical Proofs & CoT)",
      "provider": "ollama",
      "model": "apex-phi-think-14b",
      "apiBase": "http://localhost:11434"
    }
  ],
  "tabAutocompleteModel": {
    "title": "ApexPhi-Coder-14B Precision Autocomplete",
    "provider": "ollama",
    "model": "apex-phi-coder-14b",
    "apiBase": "http://localhost:11434"
  }
}

🔗 Related Resources

ApexCode-14B Model: huggingface.co/nileshc90/apex-code-14b-instruct
ApexThink-14B Model: huggingface.co/nileshc90/apex-think-14b
ApexMistral-22B Model: huggingface.co/nileshc90/apex-mistral-22b
GitHub Repository: github.com/nileshchakraborty/apex-phi-14b

Downloads last month: -; Downloads are not tracked for this model. How to track

Model tree for nileshc90/apex-phi-14b

Base model

microsoft/phi-4

Finetuned

(130)

this model