Instructions to use jkminder/d26_973m_seed2_sft with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use jkminder/d26_973m_seed2_sft with Transformers:

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

pipe = pipeline("text-generation", model="jkminder/d26_973m_seed2_sft", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("jkminder/d26_973m_seed2_sft", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use jkminder/d26_973m_seed2_sft with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "jkminder/d26_973m_seed2_sft"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jkminder/d26_973m_seed2_sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/jkminder/d26_973m_seed2_sft

SGLang

How to use jkminder/d26_973m_seed2_sft 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 "jkminder/d26_973m_seed2_sft" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jkminder/d26_973m_seed2_sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "jkminder/d26_973m_seed2_sft" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jkminder/d26_973m_seed2_sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use jkminder/d26_973m_seed2_sft with Docker Model Runner:
```
docker model run hf.co/jkminder/d26_973m_seed2_sft
```

jkminder commited on 24 days ago

Commit

f0baf65

verified ·

1 Parent(s): 8548da3

Scaling Ladder d26_973m_seed2_sft main: card refresh

Browse files

Files changed (1) hide show

README.md +33 -3

README.md CHANGED Viewed

@@ -69,6 +69,34 @@ same harness across all runs and sizes). "SFT val bpb" is the run's final
 validation loss (bits per byte) on the mixture's held-out split. A "-"
 means that run's eval has not landed yet.
 ## Usage
 The chat template is bundled; format conversations with
@@ -95,9 +123,11 @@ transformers class, so the modeling code ships in the repository
 `<|assistant_end|>`; sampling defaults (temperature 0.6, top_k 50) ship in
 `generation_config.json`. The template renders nanochat's chat format
 token-for-token (a leading system message is merged into the first user
-message); conversion is verified per revision by chat-template, logit and
-loss equivalence against the original training code
-(`verify_results.json`, where present).
 ## Architecture, tokenizer, training data

 validation loss (bits per byte) on the mixture's held-out split. A "-"
 means that run's eval has not landed yet.
+## Anneal-mark chat-SFTs
+The base repository also holds the model annealed at every mark of its
+pretraining run: base revision `TPP_x` is the model annealed at x tokens per
+parameter (`TPP_200` is the base `main`). The revisions below apply the same
+chat-SFT recipe to each of those marks, one run per mark (SFT data seed 0,
+replicate 1). Their names carry the mark with three digits (`TPP_010` ..
+`TPP_180`); each row links the base revision it was trained from. These runs
+were trained at code commit `a06bf32`, which clamps the SFT
+learning-rate schedule so the last step cannot run at a negative learning
+rate; the TPP_200 chat-SFTs above predate that fix, and their one final step
+ran at a slightly negative learning rate (about -0.0004x the peak, read from
+a d12 run's log; the factor depends on the step count).
+| revision | base revision (annealed at) | base step | SFT step | SFT val bpb | trained on |
+|---|---|---|---|---|---|
+| TPP_010 | [`TPP_10`](https://huggingface.co/jkminder/d26_973m_seed2/tree/TPP_10) (10 tokens per parameter) | 8477 | 467 | 0.2829 | squirtle |
+| TPP_020 | [`TPP_20`](https://huggingface.co/jkminder/d26_973m_seed2/tree/TPP_20) (20 tokens per parameter) | 17477 | 467 | 0.2748 | bulbasaur |
+| TPP_030 | [`TPP_30`](https://huggingface.co/jkminder/d26_973m_seed2/tree/TPP_30) (30 tokens per parameter) | 25977 | 467 | 0.2726 | squirtle |
+| TPP_040 | [`TPP_40`](https://huggingface.co/jkminder/d26_973m_seed2/tree/TPP_40) (40 tokens per parameter) | 34977 | 467 | 0.2713 | bulbasaur |
+| TPP_060 | [`TPP_60`](https://huggingface.co/jkminder/d26_973m_seed2/tree/TPP_60) (60 tokens per parameter) | 52477 | 467 | 0.2694 | squirtle |
+| TPP_080 | [`TPP_80`](https://huggingface.co/jkminder/d26_973m_seed2/tree/TPP_80) (80 tokens per parameter) | 69977 | 467 | 0.2681 | squirtle |
+| TPP_100 | [`TPP_100`](https://huggingface.co/jkminder/d26_973m_seed2/tree/TPP_100) (100 tokens per parameter) | 87477 | 467 | 0.2668 | bulbasaur |
+| TPP_120 | [`TPP_120`](https://huggingface.co/jkminder/d26_973m_seed2/tree/TPP_120) (120 tokens per parameter) | 104977 | 467 | 0.2656 | bulbasaur |
+| TPP_140 | [`TPP_140`](https://huggingface.co/jkminder/d26_973m_seed2/tree/TPP_140) (140 tokens per parameter) | 122477 | 467 | 0.2644 | bulbasaur |
+| TPP_160 | [`TPP_160`](https://huggingface.co/jkminder/d26_973m_seed2/tree/TPP_160) (160 tokens per parameter) | 139977 | 467 | 0.2638 | bulbasaur |
+| TPP_180 | [`TPP_180`](https://huggingface.co/jkminder/d26_973m_seed2/tree/TPP_180) (180 tokens per parameter) | 157477 | 467 | 0.2642 | bulbasaur |
 ## Usage
 The chat template is bundled; format conversations with
 `<|assistant_end|>`; sampling defaults (temperature 0.6, top_k 50) ship in
 `generation_config.json`. The template renders nanochat's chat format
 token-for-token (a leading system message is merged into the first user
+message). Every revision's upload is byte-verified against the converted
+checkpoint (hub listing sizes and content hashes); the conversion itself is
+verified on at least one revision per repository by chat-template, logit
+and loss equivalence against the original training code — a revision that
+was verified carries the record `verify_results.json`.
 ## Architecture, tokenizer, training data