SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2101.00190 · ACL · 2021

Prefix-Tuning: Optimizing Continuous Prompts for Generation

Percy Liang, Li, Lisa Xiang

arXiv · PDF · Open in the Atlas

Code that ran

4 of 6 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (2 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 2 have no contract verdict; 2 failures of our instrument; 2 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2101.00190.json · badge: /api/ran/2101.00190.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
thudm/swissarmytransformer community
Papers with Code archive
4 ran · 2 no recorded run
FunctionStatusWhere it lives
BaseMixin Ran
ran on a synthesized input · constructed an object; no result was computed
thudm/swissarmytransformer/sat/model/finetune/prompt_tuning.py
code served (permissive licence) · get_code("6b7c651bf334e2be")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, BaseMixin: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

ARGS = {
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type BaseMixin — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a BaseMixin output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
PrefixTuningMixin Ran
ran on a synthesized input · constructed an object; no result was computed
thudm/swissarmytransformer/sat/model/finetune/prompt_tuning.py
code served (permissive licence) · get_code("e5740cff266d5a5d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, PrefixTuningMixin: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import math

torch.manual_seed(0)

class BaseMixin(torch.nn.Module):
    def __init__(self):
        super(BaseMixin, self).__init__()

def non_conflict(func):
    """mark a hook function as non-conflict,
    so that it can be compatible with any already defined hooks.
    e.g. PrefixTuningMixin.attention_fn
    """
    func.non_conflict = True
    return func

def replacable(func):
    """mark a hook function as replacable,
    so that it can be replaced by mixins added after it.
    e.g. FP32AttentionMixin.attention_fn
    """
    func.replacable = True
    return func

class PrefixTuningMixin(BaseMixin):

    def __init__(self, num_layers, hidden_size_per_attention_head, num_attention_heads, prefix_len):
        super().__init__()
        self.prefix = torch.nn.ParameterList([torch.nn.Parameter(torch.randn(2, num_attention_heads, prefix_len, hidden_size_per_attention_head) * 0.01) for layer_id in range(num_layers)])
        self.prefix_len = prefix_len

    @non_conflict
    def attention_fn(self, q, k, v, mask, dropout_fn, old_impl=None, **kw_args):
        prefix_k, prefix_v = self.prefix[kw_args['layer_id']]
        b, nh, seq_len, hidden_size = k.shape
        prefix_k = prefix_k.unsqueeze(0).expand(b, nh, -1, hidden_size)
        prefix_v = prefix_v.unsqueeze(0).expand(b, nh, -1, hidden_size)
        k = torch.cat((k, prefix_k), dim=2)
        v = torch.cat((v, prefix_v), dim=2)
        if mask.numel() > 1:
            mask_prefixed = torch.ones(self.prefix_len, device=mask.device, dtype=mask.dtype)
            mask_prefixed = mask_prefixed.expand(*mask.size()[:-1], -1)
            mask = torch.cat((mask, mask_prefixed), dim=-1)
        return old_impl(q, k, v, mask, dropout_fn, **kw_args)

ARGS = {
    "num_layers": 2,
    "hidden_size_per_attention_head": 4,
    "num_attention_heads": 3,
    "prefix_len": 5
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PrefixTuningMixin — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a PrefixTuningMixin output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
non_conflict Ran
ran; our contract draft was wrong, not the code
thudm/swissarmytransformer/sat/model/finetune/prompt_tuning.py
code served (permissive licence) · get_code("3823a93a9e064f82")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned function. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

def sample_hook_function(x):
    """A sample function to be marked as non-conflict"""
    return x + 1

ARGS = {
    "func": sample_hook_function
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
replacable Ran
ran; our contract draft was wrong, not the code
thudm/swissarmytransformer/sat/model/finetune/prompt_tuning.py
code served (permissive licence) · get_code("7ede45f27f0943d8")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned function. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

def sample_hook_function(x):
    """A simple hook function to be marked as replacable"""
    return x * 2

ARGS = {
    "func": sample_hook_function
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
attention_fn_default Not yet run
unverified (harvested, no recorded run)
thudm/swissarmytransformer/sat/model/finetune/prompt_tuning.py
code served (permissive licence) · get_code("844b89380ab187e6")
How it ran

no invocation recorded.

standard_attention Not yet run
unverified (harvested, no recorded run)
thudm/swissarmytransformer/sat/model/finetune/prompt_tuning.py
code served (permissive licence) · get_code("82481205cb7b429d")
How it ran

no invocation recorded.

Also present in other papers' code

Of 6 samples lifted from this paper's linked repositories, 6 carry a function identity and 0 do not. Of the 6 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 6 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Fine-tuning is the de facto way of leveraging large pretrained language models for downstream tasks. However, fine-tuning modifies all the language model parameters and therefore necessitates storing a full copy for each task. In this paper, we propose prefix-tuning, a lightweight alternative to fine-tuning for natural language generation tasks, which keeps language model parameters frozen and instead optimizes a sequence of continuous task-specific vectors, which we call the prefix. Prefix-tuning draws inspiration from prompting for language models, allowing subsequent tokens to attend to this prefix as if it were "virtual tokens". We apply prefix-tuning to GPT-2 for table-totext generation and to BART for summarization. We show that by modifying only 0.1% of the parameters, prefix-tuning obtains comparable performance in the full data setting, outperforms fine-tuning in low-data settings, and extrapolates better to examples with topics that are unseen during training.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2101.00190")
get_code_for_paper("2101.00190")
have("2101.00190")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2101.00190.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2101.00190.svg)](https://syntology.ai/paper/2101.00190)

Connect an agent — have() is free.