GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
pinkelephantlimited
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
pinkelephantlimited
Model Tree
Input Modalities
Output Modalities
Supported Functionality
| Parameter | Value |
|---|---|
| Architecture | LLaMA (decoder-only) with GQA |
| Parameters | 88,945,152 |
| Hidden size | 768 |
| Intermediate size | 2,048 |
| Num layers | 8 |
| Num attention heads | 12 |
| Num KV heads | 4 (GQA) |
| Max sequence length | 1,024 |
| Vocabulary size | 50,257 (GPT-2) |
| Position encoding | RoPE |
| Activation | SwiGLU |
| Normalization | Pre-RMSNorm |
| Weight tying | Yes (embedding ↔ lm_head) |
| Precision | fp32 |
| Source | Type | Size |
|---|---|---|
| FineWeb-Edu (sample-10BT) | Educational web text | 150,000 examples |
| Parameter | Value |
|---|---|
| Hardware | 15GB VRAM |
| Optimizer | AdamW |
| Learning rate | 2e-4 |
| Batch size | 8 |
| Gradient accumulation | 4 |
| Effective batch size | 32 |
| Warmup steps | 500 |
| Weight decay | 0.01 |
| Training time | ~45 minutes |
python
from transformers import pipelinepipe = pipeline("text-generation", model="pinkelephantlimited/pink-elephant-90m")output = pipe("5 tips for better sleep:", max_new_tokens=100)[0]["generated_text"]print(output)
This model was trained entirely from scratch on permissively licensed data. No fine-tuning, no transfer learning — 100% original weights.
MIT
Grouped-Query Attention makes this efficient on mobile GPUs:
python
from transformers import pipelinepipe = pipeline("text-generation", model="pinkelephantlimited/pink-elephant-90m")# Social media post draftspipe("5 tips for better sleep: 1. Stick to a schedule", max_new_tokens=100)pipe("Breaking: Scientists have discovered", max_new_tokens=80)
python
# Draft intros and outlines for content creatorspipe("The rise of remote work has fundamentally changed", max_new_tokens=100)
python
# Use hidden states as features for classificationfrom transformers import AutoModelmodel = AutoModel.from_pretrained("pinkelephantlimited/pink-elephant-90m")# Extract embeddings for text classification, clustering, or search
Why 90M? GQA provides 3x faster inference than standard attention. Balances quality and speed — runs on phone GPUs and cloud instances.