Files
skills/public/ml-unsloth-finetuning.html
zeekayandHanzo Dev 9e10c1c10e Each llms index points at the other two
hanzoskills.com/llms.txt now opens with the install line, the API
reference index and the company index, so an agent that lands on any
one of the three can find the rest.

robots.txt claimed 537 documents while llms.txt counts 920. The number
is gone from robots.txt; llms.txt derives it.

public/ is regenerated from scripts/site.py, which it had fallen
behind: the copy-prompt menu and agent bar were in the generator and
not in the committed pages. tests/site.mjs passes against the result.

Co-authored-by: Hanzo Dev <[email protected]>
2026-09-11 00:27:59 -07:00

605 lines
28 KiB
HTML

<!doctype html>
<html lang="en"><head>
<meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1">
<title>ml-unsloth-finetuning</title>
<meta name="color-scheme" content="dark">
<style>
/* Committed to one visual world: a dark instrument panel. No light variant —
the dot field and the hairline rules only read on black, and a naive
inversion would wash both out. Every colour is painted explicitly so the page
never borrows a host background. */
@font-face{font-family:'Zen';src:url('/fonts/Zen-Variable.woff2') format('woff2');font-weight:100 900;font-style:normal;font-display:swap}
@font-face{font-family:'Zen';src:url('/fonts/Zen-Italic-Variable.woff2') format('woff2');font-weight:100 900;font-style:italic;font-display:swap}
@font-face{font-family:'Zen Mono';src:url('/fonts/ZenMono-Variable.woff2') format('woff2');font-weight:100 900;font-style:normal;font-display:swap}
@font-face{font-family:'Zen Mono';src:url('/fonts/ZenMono-Italic-Variable.woff2') format('woff2');font-weight:100 900;font-style:italic;font-display:swap}
@font-face{font-family:'Zen Pixel Square';src:url('/fonts/ZenPixel-Square.woff2') format('woff2');font-style:normal;font-display:swap}
:root{
--ink:#f4f4f2; --soft:#9c9c98; --faint:#6a6a66; --ground:#000; --panel:#0a0a0a;
--line:#232323; --line-hi:#3a3a3a; --accent:#f4f4f2;
--font-zen-sans:'Zen',-apple-system,BlinkMacSystemFont,"Segoe UI",Helvetica,Arial,sans-serif;
--font-zen-mono:'Zen Mono',ui-monospace,"SF Mono",SFMono-Regular,Menlo,Consolas,monospace;
--font-zen-pixel-square:'Zen Pixel Square',monospace;
--mono:var(--font-zen-mono);
--sans:var(--font-zen-sans);
/* Controls are pills and surfaces are soft, as on hanzo.ai. */
--r:999px; --r-soft:14px;
}
*{box-sizing:border-box}
html{-webkit-text-size-adjust:100%}
body{margin:0;background:var(--ground);color:var(--ink);font-family:var(--sans);
font-size:16px;line-height:1.65;-webkit-font-smoothing:antialiased}
.wrap{max-width:60rem;margin:0 auto;padding:0 1.75rem}
/* The dot field. A halftone made of one repeating radial-gradient, faded out
along the diagonal by a mask so it reads as a wave rather than wallpaper. */
/* No overflow clip here: the dot field is inset to the box already, and a clip
would cut off the copy-prompt menu. */
.hero{position:relative;border-bottom:1px solid var(--line)}
.hero::before{content:"";position:absolute;inset:0;
background-image:radial-gradient(var(--line-hi) 1px, transparent 1px);
background-size:7px 7px;
-webkit-mask-image:linear-gradient(115deg,#000 0%,rgba(0,0,0,.55) 38%,transparent 72%);
mask-image:linear-gradient(115deg,#000 0%,rgba(0,0,0,.55) 38%,transparent 72%);
pointer-events:none}
.hero .wrap{position:relative;padding-top:5.5rem;padding-bottom:4rem}
nav{border-bottom:1px solid var(--line)}
nav .in{max-width:60rem;margin:0 auto;padding:1.1rem 1.75rem;display:flex;
justify-content:space-between;align-items:center;gap:1rem}
.brand{font-family:var(--mono);font-size:.85rem;letter-spacing:.14em;
text-transform:uppercase;color:var(--ink);text-decoration:none;
display:inline-flex;align-items:center;gap:.6rem}
.brand i{font-style:normal;color:var(--faint)}
.mark{width:1.15em;height:1.15em;display:block;flex:0 0 auto}
nav a.up{font-family:var(--mono);font-size:.72rem;letter-spacing:.14em;
text-transform:uppercase;color:var(--soft);text-decoration:none}
nav a.up:hover{color:var(--ink)}
h1{font-family:var(--mono);font-size:clamp(2rem,5.4vw,3.4rem);line-height:1.12;
letter-spacing:-.02em;margin:0 0 1.1rem;font-weight:500;text-wrap:balance}
h1 em{font-style:normal;color:var(--faint)}
.lede{color:var(--soft);font-size:1.05rem;max-width:52ch;margin:0 0 2.25rem}
.btns{display:flex;flex-wrap:wrap;gap:.7rem;margin:0 0 .5rem;align-items:center}
.btn{display:inline-flex;align-items:center;gap:.5rem;font-family:var(--mono);
font-size:.72rem;letter-spacing:.14em;text-transform:uppercase;
padding:.8rem 1.4rem;border:1px solid var(--line-hi);border-radius:var(--r);
color:var(--ink);text-decoration:none;background:transparent;cursor:pointer;
line-height:1.2;font-weight:400}
.btn:hover{background:#141414;text-decoration:none;border-color:var(--soft)}
.btn.solid{background:var(--ink);color:#000;border-color:var(--ink)}
.btn.solid:hover{background:#d8d8d4}
.btn .chev{width:.7em;height:.7em;flex:0 0 auto;transition:transform .15s ease}
.btn[aria-expanded="true"] .chev{transform:rotate(180deg)}
/* Copy-prompt selector. Each row copies the prompt for its own target. */
.copy{position:relative;display:inline-flex}
.menu{position:absolute;top:calc(100% + .5rem);left:0;z-index:20;min-width:21rem;
background:var(--panel);border:1px solid var(--line-hi);border-radius:var(--r-soft);
padding:.4rem;display:flex;flex-direction:column;
box-shadow:0 12px 32px rgba(0,0,0,.75)}
.menu[hidden]{display:none}
.mi{display:flex;align-items:center;gap:.7rem;width:100%;text-align:left;
padding:.62rem .8rem;border:0;border-radius:calc(var(--r-soft) - .35rem);
background:transparent;color:var(--ink);font-family:var(--mono);font-size:.76rem;
letter-spacing:.04em;text-decoration:none;cursor:pointer;line-height:1.3;
white-space:nowrap}
.mi:hover,.mi:focus-visible{background:#181818;outline:none}
.mi .mark{width:1.05rem;height:1.05rem;color:var(--soft)}
.mi:hover .mark{color:var(--ink)}
.mi u{margin-left:auto;color:var(--faint);text-decoration:none;font-size:.68rem;
letter-spacing:.1em;text-transform:uppercase}
/* Works-with bar: one scrolling row, name and mark per agent. */
.strip{border-bottom:1px solid var(--line);background:var(--panel)}
.strip .in{max-width:60rem;margin:0 auto;padding:.9rem 1.75rem;display:flex;
align-items:center;gap:1.25rem;font-family:var(--mono);font-size:.72rem;
letter-spacing:.1em;text-transform:uppercase;color:var(--faint)}
.strip b{color:var(--soft);font-weight:400;flex:0 0 auto;
border-right:1px solid var(--line);padding-right:1.25rem;align-self:stretch;
display:flex;align-items:center}
.bar{display:flex;align-items:center;gap:.5rem;flex:1 1 auto;min-width:0;
overflow-x:auto;scrollbar-width:none;scroll-snap-type:x proximity;
-webkit-overflow-scrolling:touch;
-webkit-mask-image:linear-gradient(90deg,#000 calc(100% - 3rem),transparent);
mask-image:linear-gradient(90deg,#000 calc(100% - 3rem),transparent)}
.bar::-webkit-scrollbar{display:none}
.agent{display:inline-flex;align-items:center;gap:.45rem;flex:0 0 auto;
padding:.4rem .85rem;border:1px solid var(--line);border-radius:var(--r);
color:var(--soft);white-space:nowrap;scroll-snap-align:start;
text-decoration:none;background:var(--ground)}
.agent:hover{color:var(--ink);border-color:var(--line-hi);text-decoration:none}
.agent .mark{width:.95rem;height:.95rem}
h2{font-family:var(--mono);font-size:1.1rem;letter-spacing:-.01em;margin:2.75rem 0 .75rem;font-weight:500}
h3{font-size:1rem;margin:2rem 0 .5rem;font-weight:600}
h4,h5,h6{font-size:.94rem;margin:1.5rem 0 .5rem;font-weight:600}
p{margin:0 0 1rem;max-width:70ch}
ul,ol{margin:0 0 1rem;padding-left:1.35rem;max-width:70ch}
li{margin:.22rem 0}
a{color:var(--ink);text-decoration:underline;text-underline-offset:3px;
text-decoration-color:var(--line-hi)}
a:hover{text-decoration-color:var(--ink)}
a:focus-visible{outline:1px solid var(--ink);outline-offset:3px}
code{font-family:var(--mono);font-size:.85em;background:var(--panel);
border:1px solid var(--line);border-radius:5px;padding:.1em .36em}
pre{background:var(--panel);border:1px solid var(--line);padding:1rem 1.1rem;
border-radius:var(--r-soft);overflow-x:auto;margin:0 0 1.25rem}
pre code{background:none;border:0;padding:0;font-size:.82rem;line-height:1.7}
blockquote{margin:0 0 1rem;padding-left:1rem;border-left:1px solid var(--line-hi);color:var(--soft)}
hr{border:0;border-top:1px solid var(--line);margin:2.5rem 0}
table{border-collapse:collapse;width:100%;margin:0 0 1.25rem;font-size:.92rem;
display:block;overflow-x:auto}
th,td{border:1px solid var(--line);padding:.5rem .7rem;text-align:left}
.curl{display:block;background:var(--panel);border:1px solid var(--line);
border-radius:var(--r-soft);padding:.95rem 1.1rem;font-family:var(--mono);
font-size:.8rem;color:var(--soft);overflow-x:auto;margin:0 0 2.5rem;
white-space:pre;line-height:1.9}
.curl b{color:var(--ink);font-weight:400}
.curl i{font-style:normal;color:var(--faint)}
/* index */
#q{width:100%;padding:.85rem 1.2rem;font-family:var(--mono);font-size:.85rem;color:var(--ink);
background:var(--panel);border:1px solid var(--line);border-radius:var(--r);margin:0 0 .6rem}
#q::placeholder{color:var(--faint);text-transform:uppercase;letter-spacing:.1em;font-size:.72rem}
#q:focus{outline:none;border-color:var(--soft)}
#count{color:var(--faint);font-family:var(--mono);font-size:.72rem;letter-spacing:.1em;
text-transform:uppercase;margin:0 0 2rem}
.cat{margin:2.5rem 0 .5rem;font-family:var(--mono);font-size:.7rem;letter-spacing:.16em;
text-transform:uppercase;color:var(--faint);border-bottom:1px solid var(--line);
padding-bottom:.45rem}
.row{display:block;padding:.6rem .9rem;margin:0 -.9rem;border-bottom:1px solid var(--line);
border-radius:var(--r-soft);text-decoration:none}
.row:hover{background:#0a0a0a}
.nm{font-family:var(--mono);font-size:.84rem;color:var(--ink);display:block}
.ds{color:var(--soft);font-size:.9rem;display:block;margin-top:.12rem}
footer{margin-top:4rem;padding:2rem 0 4rem;border-top:1px solid var(--line);
color:var(--faint);font-size:.85rem}
footer a{color:var(--soft)}
.fmap{display:grid;gap:1.75rem;grid-template-columns:repeat(auto-fit,minmax(11rem,1fr));
margin:0 0 2rem;font-size:.85rem}
.fmap b{display:block;font-family:var(--mono);font-size:.68rem;letter-spacing:.16em;
text-transform:uppercase;color:var(--faint);font-weight:400;margin:0 0 .6rem;
border-bottom:1px solid var(--line);padding-bottom:.4rem}
.fmap a{display:inline-block;color:var(--ink);text-decoration:none;margin:.3rem .35rem 0 0}
.fmap a:hover{text-decoration:underline}
.fmap div{color:var(--faint);line-height:1.5}
.fend{border-top:1px solid var(--line);padding-top:1.25rem;font-size:.82rem}
@media (max-width:640px){
.hero .wrap{padding-top:3.25rem;padding-bottom:2.75rem}
.btns{gap:.5rem}
.btn{padding:.75rem 1.15rem;font-size:.68rem}
.copy,.copy .btn{width:100%;justify-content:center}
.menu{left:0;right:0;min-width:0}
.strip .in{flex-direction:column;align-items:flex-start;gap:.6rem;padding:.9rem 1.75rem}
.strip b{border-right:0;padding-right:0}
.bar{width:100%}
}
</style>
<link rel="alternate" type="text/markdown" href="/ml-unsloth-finetuning.md">
<link rel="icon" href="/favicon.svg">
<link rel="canonical" href="https://hanzoskills.com/ml-unsloth-finetuning">
</head>
<!-- Reading this with a machine? Every page here is also plain markdown.
This one: https://hanzoskills.com/ml-unsloth-finetuning.md
Index: https://hanzoskills.com/llms.txt
Corpus: https://hanzoskills.com/all.md
Append .md to any path for the markdown twin. -->
<body>
<nav><div class="in">
<a class="brand" href="/"><svg class="mark" viewBox="0 0 67 67" fill="currentColor" aria-hidden="true"><path d="M22.21 67V44.6369H0V67H22.21Z"/><path d="M66.7038 22.3184H22.2534L0.0878906 44.6367H44.4634L66.7038 22.3184Z"/><path d="M22.21 0H0V22.3184H22.21V0Z"/><path d="M66.7198 0H44.5098V22.3184H66.7198V0Z"/><path d="M66.7198 67V44.6369H44.5098V67H66.7198Z"/></svg> HANZO<i>SKILLS</i></a>
<a class="up" href="/">All skills</a>
</div></nav>
<div class="wrap">
<h1>ml-unsloth-finetuning</h1><p class="lede">Fine-tuning LLMs (Llama, Mistral, Qwen) efficiently</p><h1>Unsloth Fine-Tuning</h1>
<p><strong>Scope</strong>: Fast LLM fine-tuning with Unsloth, memory optimization, multi-GPU, Flash Attention <strong>Lines</strong>: ~350 <strong>Last Updated</strong>: 2025-10-18</p>
<h2>When to Use This Skill</h2>
<p>Activate this skill when:</p>
<ul><li>Fine-tuning LLMs (Llama, Mistral, Qwen) efficiently</li><li>Working in memory-constrained GPU environments</li><li>Needing 2-5x faster training than standard methods</li><li>Optimizing training costs with reduced memory usage</li><li>Requiring rapid iteration on fine-tuning experiments</li><li>Training with consumer GPUs (24GB VRAM or less)</li></ul>
<h2>Core Concepts</h2>
<h3>Unsloth Architecture</h3>
<p><strong>Speed Optimizations</strong>:</p>
<ul><li>Flash Attention 2 integration for 2x faster attention</li><li>Custom CUDA kernels for backward pass optimization</li><li>Optimized RoPE (Rotary Position Embeddings) computation</li><li>Fused layernorm and cross-entropy operations</li><li>4-bit quantization with minimal accuracy loss</li></ul>
<p><strong>Memory Efficiency</strong>:</p>
<ul><li>Reduces VRAM usage by 30-60% vs standard training</li><li>Enables larger batch sizes on same hardware</li><li>Supports gradient checkpointing out of the box</li><li>Efficient mixed precision (bfloat16) handling</li></ul>
<p><strong>Compatibility</strong>:</p>
<ul><li>Works with PEFT/LoRA for parameter-efficient training</li><li>Supports Hugging Face Transformers ecosystem</li><li>Compatible with Modal.com GPU workloads</li><li>Drop-in replacement for standard fine-tuning</li></ul>
<h3>Supported Models</h3>
<p><strong>Llama Family</strong>:</p>
<ul><li>Llama 3.2 (1B, 3B), Llama 3.1 (8B, 70B, 405B)</li><li>Llama 3 (8B, 70B), Llama 2 (7B, 13B, 70B)</li></ul>
<p><strong>Mistral Family</strong>:</p>
<ul><li>Mistral 7B v0.3, Mixtral 8x7B, 8x22B</li><li>Mistral Small, Mistral NeMo</li></ul>
<p><strong>Other Models</strong>:</p>
<ul><li>Qwen 2.5 (0.5B - 72B), Qwen 2 (0.5B - 72B)</li><li>Gemma 2 (2B, 9B, 27B), Phi-3.5 (mini)</li><li>TinyLlama, OpenELM variants</li></ul>
<h3>Quantization Strategies</h3>
<p><strong>4-bit QLoRA</strong> (Recommended):</p>
<ul><li>Uses NF4 (Normal Float 4) quantization</li><li>Maintains 99%+ of full precision performance</li><li>Reduces memory by 4x vs FP16</li><li>Allows fine-tuning 70B models on 48GB GPU</li></ul>
<p><strong>16-bit Training</strong>:</p>
<ul><li>Full bfloat16 precision for maximum quality</li><li>Requires 2-4x more VRAM</li><li>Recommended for models &lt;13B with sufficient VRAM</li></ul>
<hr>
<h2>Patterns</h2>
<h3>Basic Llama 3 Fine-Tuning</h3>
<pre><code class="lang-python">from unsloth import FastLanguageModel
import torch
# Load model with 4-bit quantization
model, tokenizer = FastLanguageModel.from_pretrained(
model_name=&quot;unsloth/llama-3-8b-bnb-4bit&quot;,
max_seq_length=2048,
dtype=None, # Auto-detect (bfloat16 for Ampere+)
load_in_4bit=True,
)
# Configure LoRA
model = FastLanguageModel.get_peft_model(
model,
r=16, # LoRA rank
target_modules=[
&quot;q_proj&quot;, &quot;k_proj&quot;, &quot;v_proj&quot;, &quot;o_proj&quot;,
&quot;gate_proj&quot;, &quot;up_proj&quot;, &quot;down_proj&quot;
],
lora_alpha=16,
lora_dropout=0, # Optimized for 0 dropout
bias=&quot;none&quot;,
use_gradient_checkpointing=&quot;unsloth&quot;, # Unsloth&#x27;s optimized checkpointing
random_state=3407,
)</code></pre>
<h3>Dataset Preparation</h3>
<pre><code class="lang-python">from datasets import load_dataset
# Load dataset (Alpaca format)
dataset = load_dataset(&quot;yahma/alpaca-cleaned&quot;, split=&quot;train&quot;)
# Format for instruction tuning
alpaca_prompt = &quot;&quot;&quot;Below is an instruction that describes a task. Write a response that appropriately completes the request.
### Instruction:
{}
### Response:
{}&quot;&quot;&quot;
def format_prompts(examples):
instructions = examples[&quot;instruction&quot;]
outputs = examples[&quot;output&quot;]
texts = []
for instruction, output in zip(instructions, outputs):
text = alpaca_prompt.format(instruction, output) + tokenizer.eos_token
texts.append(text)
return {&quot;text&quot;: texts}
# Apply formatting
dataset = dataset.map(format_prompts, batched=True)</code></pre>
<h3>Training with SFTTrainer</h3>
<pre><code class="lang-python">from trl import SFTTrainer
from transformers import TrainingArguments
trainer = SFTTrainer(
model=model,
tokenizer=tokenizer,
train_dataset=dataset,
dataset_text_field=&quot;text&quot;,
max_seq_length=2048,
dataset_num_proc=2,
packing=False, # Can make training 5x faster for short sequences
args=TrainingArguments(
per_device_train_batch_size=2,
gradient_accumulation_steps=4, # Effective batch size = 8
warmup_steps=5,
max_steps=60, # Or num_train_epochs=1
learning_rate=2e-4,
fp16=not torch.cuda.is_bf16_supported(),
bf16=torch.cuda.is_bf16_supported(),
logging_steps=1,
optim=&quot;adamw_8bit&quot;, # Reduced memory optimizer
weight_decay=0.01,
lr_scheduler_type=&quot;linear&quot;,
seed=3407,
output_dir=&quot;outputs&quot;,
),
)
# Train
trainer_stats = trainer.train()</code></pre>
<h3>Multi-GPU Training</h3>
<pre><code class="lang-python"># Use with torchrun or Modal multi-GPU
import modal
app = modal.App(&quot;unsloth-multi-gpu&quot;)
image = (
modal.Image.debian_slim()
.pip_install(&quot;unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git&quot;)
.pip_install(&quot;xformers&quot;, &quot;trl&quot;, &quot;peft&quot;, &quot;accelerate&quot;)
)
@app.function(
gpu=&quot;a100:2&quot;, # 2x A100
image=image,
timeout=7200
)
def multi_gpu_training():
from unsloth import FastLanguageModel
from accelerate import Accelerator
accelerator = Accelerator()
# Model automatically distributed across GPUs
model, tokenizer = FastLanguageModel.from_pretrained(
model_name=&quot;unsloth/llama-3-70b-bnb-4bit&quot;,
max_seq_length=4096,
dtype=None,
load_in_4bit=True,
device_map=&quot;auto&quot; # Auto-distribute across GPUs
)
model = FastLanguageModel.get_peft_model(
model,
r=32, # Higher rank for 70B model
target_modules=[&quot;q_proj&quot;, &quot;k_proj&quot;, &quot;v_proj&quot;, &quot;o_proj&quot;],
lora_alpha=32,
)
# Train with accelerate
# (SFTTrainer handles multi-GPU automatically)</code></pre>
<h3>Saving and Loading</h3>
<pre><code class="lang-python"># Save LoRA adapter only (small, ~100MB)
model.save_pretrained(&quot;lora_model&quot;)
tokenizer.save_pretrained(&quot;lora_model&quot;)
# Save merged model (full weights)
model.save_pretrained_merged(
&quot;merged_model&quot;,
tokenizer,
save_method=&quot;merged_16bit&quot; # or &quot;lora&quot;, &quot;merged_4bit&quot;
)
# Push to Hugging Face Hub
model.push_to_hub_merged(
&quot;username/model-name&quot;,
tokenizer,
save_method=&quot;merged_16bit&quot;,
token=&quot;hf_...&quot;
)
# Load for inference
model, tokenizer = FastLanguageModel.from_pretrained(
model_name=&quot;lora_model&quot;,
max_seq_length=2048,
dtype=None,
load_in_4bit=True,
)
FastLanguageModel.for_inference(model) # Enable fast inference mode</code></pre>
<h3>Inference After Training</h3>
<pre><code class="lang-python"># Enable fast inference (disables dropout, etc.)
FastLanguageModel.for_inference(model)
# Generate
inputs = tokenizer(
[alpaca_prompt.format(
&quot;What is the capital of France?&quot;,
&quot;&quot; # Leave output empty for generation
)],
return_tensors=&quot;pt&quot;
).to(&quot;cuda&quot;)
outputs = model.generate(
**inputs,
max_new_tokens=128,
use_cache=True,
temperature=1.5,
min_p=0.1
)
print(tokenizer.batch_decode(outputs))</code></pre>
<h3>Modal.com Deployment</h3>
<pre><code class="lang-python">import modal
app = modal.App(&quot;unsloth-finetune&quot;)
image = (
modal.Image.debian_slim()
.pip_install(
&quot;unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git&quot;,
&quot;xformers&quot;,
&quot;trl&quot;,
&quot;peft&quot;
)
)
volume = modal.Volume.from_name(&quot;model-volume&quot;, create_if_missing=True)
@app.function(
gpu=&quot;l40s&quot;, # Recommended: great price/performance
image=image,
volumes={&quot;/models&quot;: volume},
timeout=3600
)
def finetune_llama(
dataset_name: str,
output_path: str = &quot;/models/finetuned&quot;
):
from unsloth import FastLanguageModel
from trl import SFTTrainer
from transformers import TrainingArguments
from datasets import load_dataset
# Load model
model, tokenizer = FastLanguageModel.from_pretrained(
model_name=&quot;unsloth/llama-3-8b-bnb-4bit&quot;,
max_seq_length=2048,
dtype=None,
load_in_4bit=True,
)
# Apply LoRA
model = FastLanguageModel.get_peft_model(
model,
r=16,
target_modules=[&quot;q_proj&quot;, &quot;k_proj&quot;, &quot;v_proj&quot;, &quot;o_proj&quot;],
lora_alpha=16,
)
# Load dataset
dataset = load_dataset(dataset_name, split=&quot;train&quot;)
# Train
trainer = SFTTrainer(
model=model,
tokenizer=tokenizer,
train_dataset=dataset,
dataset_text_field=&quot;text&quot;,
max_seq_length=2048,
args=TrainingArguments(
per_device_train_batch_size=2,
gradient_accumulation_steps=4,
warmup_steps=5,
max_steps=100,
learning_rate=2e-4,
fp16=not torch.cuda.is_bf16_supported(),
bf16=torch.cuda.is_bf16_supported(),
logging_steps=1,
optim=&quot;adamw_8bit&quot;,
output_dir=&quot;/tmp/output&quot;,
),
)
trainer.train()
# Save to volume
model.save_pretrained(output_path)
tokenizer.save_pretrained(output_path)
volume.commit()
return {&quot;status&quot;: &quot;complete&quot;, &quot;path&quot;: output_path}
@app.local_entrypoint()
def main():
result = finetune_llama.remote(&quot;yahma/alpaca-cleaned&quot;)
print(result)</code></pre>
<h3>Chat Template Fine-Tuning</h3>
<pre><code class="lang-python"># ChatML format
chat_template = &quot;&quot;&quot;&lt;|im_start|&gt;system
{system}&lt;|im_end|&gt;
&lt;|im_start|&gt;user
{user}&lt;|im_end|&gt;
&lt;|im_start|&gt;assistant
{assistant}&lt;|im_end|&gt;&quot;&quot;&quot;
def format_chat(examples):
texts = []
for system, user, assistant in zip(
examples[&quot;system&quot;],
examples[&quot;user&quot;],
examples[&quot;assistant&quot;]
):
text = chat_template.format(
system=system,
user=user,
assistant=assistant
) + tokenizer.eos_token
texts.append(text)
return {&quot;text&quot;: texts}
dataset = dataset.map(format_chat, batched=True)</code></pre>
<hr>
<h2>Quick Reference</h2>
<h3>GPU Requirements by Model Size</h3>
<pre><code>Model Size | 4-bit VRAM | 16-bit VRAM | Recommended GPU
--------------|------------|-------------|------------------
1-3B | 6-8 GB | 12-16 GB | T4, L4
7-8B | 12-16 GB | 24-32 GB | L40S
13B | 16-20 GB | 32-40 GB | L40S, A100
70B | 40-48 GB | 140+ GB | A100 80GB, H100</code></pre>
<h3>Unsloth Installation</h3>
<pre><code class="lang-bash"># Standard installation
pip install &quot;unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git&quot;
pip install xformers trl peft accelerate
# For CUDA 11.8
pip install &quot;unsloth[cu118-ampere-new] @ git+https://github.com/unslothai/unsloth.git&quot;
# For CUDA 12.1+
pip install &quot;unsloth[cu121-ampere-new] @ git+https://github.com/unslothai/unsloth.git&quot;</code></pre>
<h3>Key Hyperparameters</h3>
<pre><code class="lang-python"># LoRA Configuration
r = 16 # Rank: 8-64 (higher = more params, better quality)
lora_alpha = 16 # Usually same as r
lora_dropout = 0 # Unsloth optimized for 0
target_modules = [ # All attention + MLP for best results
&quot;q_proj&quot;, &quot;k_proj&quot;, &quot;v_proj&quot;, &quot;o_proj&quot;,
&quot;gate_proj&quot;, &quot;up_proj&quot;, &quot;down_proj&quot;
]
# Training Configuration
batch_size = 2 # Per device (increase for larger GPUs)
gradient_accumulation = 4 # Effective batch = batch_size * accumulation
learning_rate = 2e-4 # Standard for LoRA
warmup_steps = 5 # 5-10% of total steps
max_seq_length = 2048 # Model&#x27;s max (512-4096)</code></pre>
<h3>Common Commands</h3>
<pre><code class="lang-python"># Enable inference mode
FastLanguageModel.for_inference(model)
# Save adapter only
model.save_pretrained(&quot;path&quot;)
# Save merged 16-bit
model.save_pretrained_merged(&quot;path&quot;, tokenizer, save_method=&quot;merged_16bit&quot;)
# Save merged 4-bit (smallest)
model.save_pretrained_merged(&quot;path&quot;, tokenizer, save_method=&quot;merged_4bit&quot;)
# Push to Hub
model.push_to_hub_merged(&quot;user/model&quot;, tokenizer, save_method=&quot;merged_16bit&quot;)</code></pre>
<hr>
<h2>Anti-Patterns</h2>
<p>❌ <strong>Using standard Transformers for fine-tuning</strong>: Slower and uses more memory ✅ Use Unsloth for 2-5x speedup with same results</p>
<p>❌ <strong>Loading full precision models</strong>: Wastes VRAM on consumer GPUs ✅ Always use <code>load_in_4bit=True</code> for models &gt;7B</p>
<p>❌ <strong>Wrong dtype for GPU architecture</strong>:</p>
<pre><code class="lang-python"># ❌ Bad: Forces FP16 on Ampere+ GPUs
args = TrainingArguments(fp16=True, bf16=False)
# ✅ Good: Auto-detect best dtype
args = TrainingArguments(
fp16=not torch.cuda.is_bf16_supported(),
bf16=torch.cuda.is_bf16_supported()
)</code></pre>
<p>❌ <strong>Tiny batch sizes without gradient accumulation</strong>: Unstable training ✅ Use <code>gradient_accumulation_steps</code> to increase effective batch size</p>
<p>❌ <strong>Too many LoRA target modules</strong>: Wastes memory for marginal gains ✅ Start with attention layers only: <code>[&quot;q_proj&quot;, &quot;k_proj&quot;, &quot;v_proj&quot;, &quot;o_proj&quot;]</code></p>
<p>❌ <strong>Ignoring sequence length</strong>: OOM errors or wasted compute ✅ Set <code>max_seq_length</code> to actual data length (analyze dataset first)</p>
<p>❌ <strong>Not using gradient checkpointing</strong>: Runs out of VRAM ✅ Always use <code>use_gradient_checkpointing=&quot;unsloth&quot;</code> for large models</p>
<p>❌ <strong>Saving full model every epoch</strong>: Wastes disk space ✅ Save LoRA adapter only (~100MB) or push merged model once at end</p>
<p>❌ <strong>Using standard AdamW</strong>: Higher memory usage ✅ Use <code>optim=&quot;adamw_8bit&quot;</code> for 2x memory reduction</p>
<hr>
<h2>Related Skills</h2>
<ul><li><code>lora-peft-techniques.md</code> - LoRA configuration, rank selection, merging</li><li><code>llm-dataset-preparation.md</code> - Dataset formatting, quality, validation</li><li><code>modal-gpu-workloads.md</code> - GPU selection, cost optimization, deployment</li><li><code>huggingface-autotrain.md</code> - Alternative no-code fine-tuning approach</li><li><code>llm-inference-optimization.md</code> - Post-training inference optimization</li><li><code>model-evaluation.md</code> - Evaluating fine-tuned model quality</li></ul>
<hr>
<p><strong>Last Updated</strong>: 2025-10-18 <strong>Format Version</strong>: 1.0 (Atomic)</p>
<footer>
<div class="fmap">
<div><b>Build</b>
<a href="https://hanzo.app">hanzo.app</a> app builder
<a href="https://hanzo.codes">hanzo.codes</a> code with an agent
<a href="https://hanzoskills.com">hanzoskills.com</a> what an agent reads</div>
<div><b>Run</b>
<a href="https://console.hanzo.ai">console.hanzo.ai</a> cloud console
<a href="https://api.hanzo.ai">api.hanzo.ai</a> the API</div>
<div><b>Account</b>
<a href="https://hanzo.id">hanzo.id</a> sign in
<a href="https://pay.hanzo.ai">pay.hanzo.ai</a> top up
<a href="https://billing.hanzo.ai">billing.hanzo.ai</a> invoices and usage</div>
<div><b>Company</b>
<a href="https://hanzo.ai">hanzo.ai</a> canonical
<a href="https://hanzo.ventures">hanzo.ventures</a> ventures</div>
</div>
<div class="fend"><a href="/ml-unsloth-finetuning.md">This page as plain text</a> · <a href="/llms.txt">Index for agents</a></div>
</footer>
</div></body></html>