Same text, two models

SmolLM2-135M and its instruction-tuned version read exactly the same tokens; only their training differs.

Your message

Both models read exactly this

SmolLM2-135M (base)

Pretrained only: it learned to continue text.

Recorded from SmolLM2-135M

Loading the model's recorded outputs for the four example messages, as plain text and in the chat template.

About the download

See how the base model continues the same text. This uses SmolLM2-135M, a 135-million-parameter base language model from Hugging Face (2024), run from its original weights by this site's own engine.

The first time, your browser downloads about 272 MB of model files from Hugging Face. They are cached for later visits. Everything runs on your device; nothing you type is sent anywhere.

Manage downloads

Loading the recorded results

SmolLM2-135M-Instruct

The same model after fine-tuning on conversations and preference tuning.

Recorded from SmolLM2-135M-Instruct

Loading the model's recorded outputs for the four example messages, as plain text and in the chat template.

About the download

See how the instruction-tuned model answers the same text. This uses SmolLM2-135M-Instruct, the chat version of SmolLM2-135M from Hugging Face (2024): the same 135 million parameters, fine-tuned on about 460,000 example conversations and then preference-tuned, run from its original weights by this site's own engine.

The first time, your browser downloads about 272 MB of model files from Hugging Face. They are cached for later visits. Everything runs on your device; nothing you type is sent anywhere.

Manage downloads

Loading the recorded results

What the models read

Chat templatePlain text

Challenges

0 of 3 complete

Find the setting where the instruct model ends its reply without writing a single word. (not yet)
With plain text, find a message that the base model answers by writing it out again. (not yet)
Get the instruct model to answer and end its turn by itself. (not yet)

Both models take their most likely next token every time, and stop at their end-of-turn token, at the end-of-text token, or after 48 tokens. The base model is the same download as the live model, microscope, and steering labs.

Try this

With the chat template, compare the two columns: the instruct model answers and ends its turn; the base model does something strange with the tags.
Switch to plain text. The base model continues your question like a page of questions, and the instruct model ends its reply at once.

Why the end-of-turn token comes after a questionPython with the standard library

from collections import Counter, defaultdict

# A toy "instruction tuning" set in the chat layout (SmolLM2-135M-Instruct
# was tuned on about 460,000 conversations; three show the idea).
pairs = [("What is the capital of France ?", "Paris ."),
         ("What is the capital of Italy ?", "Rome ."),
         ("Name a color .", "Blue .")]

def chat(user, answer=""):
    return ["<|im_start|>", "user", "\n", *user.split(), "<|im_end|>", "\n",
            "<|im_start|>", "assistant", "\n", *answer.split()]

# Count which token follows each token and each pair of tokens.
follows = defaultdict(Counter)
for user, answer in pairs:
    seq = chat(user, answer) + ["<|im_end|>"]
    for i in range(1, len(seq)):
        follows[(seq[i - 1],)][seq[i]] += 1
        if i > 1:
            follows[(seq[i - 2], seq[i - 1])][seq[i]] += 1

def next_token(tokens):
    # Use the last two tokens if that pair was seen, else the last one.
    counts = follows.get(tuple(tokens[-2:])) or follows[(tokens[-1],)]
    word, n = counts.most_common(1)[0]
    return f"{word!r} ({n} of {sum(counts.values())})"

question = "What is the capital of Spain ?"
print("plain text    ->", next_token(question.split()))
print("chat template ->", next_token(chat(question)))  # the layout, not geography
print("after Blue .  ->", next_token(chat("Name a color .", "Blue .")))

plain text    -> '<|im_end|>' (2 of 2)
chat template -> 'Paris' (1 of 3)
after Blue .  -> '<|im_end|>' (1 of 1)