SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2307.03170 · NeurIPS · 2023

Focused Transformer: Contrastive Training for Context Scaling

Yuhuai Wu, Piotr Miłoś, Henryk Michalewski, Szymon Tworkowski, Konrad Staniszewski, Mikołaj Pacek

arXiv · PDF · Open in the Atlas

Code that ran

9 of 14 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 2 violated it and 0 have no contract verdict; 7 failures of our instrument; 5 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/2307.03170.json · badge: /api/ran/2307.03170.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
CStanKonrad/long_llama canonical (classifier label)
classifier basis: census
7 ran · 5 no recorded run
cstankonrad/long_llama canonical (classifier label)
classifier basis: classifier_2025
2 ran · 0 no recorded run
FunctionStatusWhere it lives
match_keywords Ran
ran, violated the contract we drafted
CStanKonrad/long_llama/fot_continued_pretraining/EasyLM/models/llama/convert_easylm_to_hf.py
code served (permissive licence) · get_code("cc2ba1b1b0c83f6c")
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 bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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 numpy as np

def match_keywords(string, positives, negatives):
    for positive in positives:
        if positive not in string:
            return False
    for negative in negatives:
        if negative in string:
            return False
    return True

ARGS = {
    "string": "The quick brown fox jumps over the lazy dog",
    "positives": ["quick", "fox"],
    "negatives": ["cat", "slow"]
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
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
mem_apply_update Ran
ran; our contract draft was wrong, not the code
CStanKonrad/long_llama/src/longllama_utils.py
code served (permissive licence) · get_code("43de84e441595b25")
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 LongLlamaMemCache. 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
from dataclasses import dataclass
from typing import Tuple, Optional

torch.manual_seed(0)

@dataclass
class LongLlamaMemConfig:
    positionals: bool = True
    cache_dtype: torch.dtype = torch.bfloat16
    attention_grouping: Optional[Tuple[int, int]] = None

@dataclass
class LongLlamaMemCache:
    keys: torch.FloatTensor
    values: torch.FloatTensor
    masks: torch.FloatTensor

# Create small tensors with distinct dimensions
# batch_size=2, num_heads=3, mem_length=4, embed_size_per_head=5
batch_size = 2
num_heads = 3
mem_length_prev = 4
mem_length_new = 6
embed_size_per_head = 5

prev_keys = torch.randn(batch_size, num_heads, mem_length_prev, embed_size_per_head)
prev_values = torch.randn(batch_size, num_heads, mem_length_prev, embed_size_per_head)
prev_masks = torch.randn(batch_size, 1, mem_length_prev, 1)

new_keys = torch.randn(batch_size, num_heads, mem_length_new, embed_size_per_head)
new_values = torch.randn(batch_size, num_heads, mem_length_new, embed_size_per_head)
new_masks = torch.randn(batch_size, 1, mem_length_new, 1)

prev_mem_cache = LongLlamaMemCache(keys=prev_keys, values=prev_values, masks=prev_masks)
new_mem_content = LongLlamaMemCache(keys=new_keys, values=new_values, masks=new_masks)
mem_config = LongLlamaMemConfig(positionals=True, cache_dtype=torch.bfloat16, attention_grouping=None)

ARGS = {
    "prev_mem_cache": prev_mem_cache,
    "new_mem_content": new_mem_content,
    "mem_config": mem_config
}
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 LongLlamaMemCache — a LongLlamaMemCache output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a LongLlamaMemCache 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
metrics_assign_group Ran
ran; our contract draft was wrong, not the code
CStanKonrad/long_llama/instruction_fine_tuning/utils.py
code served (permissive licence) · get_code("3544c0f9d7c6110f")
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 dict. 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

torch.manual_seed(0)

# Construct a metrics dictionary with keys containing '/' separators
metrics_dict = {
    'train/loss': 0.5,
    'train/accuracy': 0.95,
    'eval/loss': 0.3,
    'eval/accuracy/top1': 0.98,
    'eval/accuracy/top5': 0.99,
}

group = 'validation'
index = 1

ARGS = {
    'metrics_dict': metrics_dict,
    'group': group,
    'index': index,
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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_numeric_to_str Ran
ran; our contract draft was wrong, not the code
CStanKonrad/long_llama/instruction_fine_tuning/utils.py
code served (permissive licence) · get_code("5a004e0660382250")
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 dict. 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 numpy as np
from typing import Dict, Any

torch.manual_seed(0)

# Construct a metrics dictionary with mixed types
metrics_dict = {
    "loss": 0.5,  # float
    "accuracy": 95,  # int
    "model_name": "ResNet50",  # str (non-numeric)
    "timestamp": np.datetime64('2024-01-01'),  # non-numeric
    "epoch": 42,  # int
    "learning_rate": 0.001,  # float
    "tensor_value": torch.tensor([1.0, 2.0, 3.0]),  # torch tensor (non-numeric)
    "config": {"hidden_dim": 128},  # dict (non-numeric)
}

ARGS = {
    "metrics_dict": metrics_dict
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
precompute_freqs_cis Ran
ran, violated the contract we drafted
CStanKonrad/long_llama/fot_continued_pretraining/EasyLM/models/llama/llama_model.py
code served (permissive licence) · get_code("1d636c739a91d422")
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 ArrayImpl, ndim 2, dtype family float, shape (4, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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 numpy as np
import jax.numpy as jnp

# Define ARGS with valid arguments for precompute_freqs_cis
ARGS = {
    "dim": 8,
    "end": 4,
    "theta": 10000.0,
    "dtype": jnp.float32
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 2, dtype family float, shape (4, 4) — shape (4, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
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
read_json Ran
ran; our contract draft was wrong, not the code
CStanKonrad/long_llama/fot_continued_pretraining/EasyLM/models/llama/convert_easylm_to_hf.py
code served (permissive licence) · get_code("c5bcf01d18bba63d")
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) and returned dict. 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 json
import tempfile
import os

# Create a temporary JSON file with test data
temp_dir = tempfile.mkdtemp()
test_json_path = os.path.join(temp_dir, "test.json")

test_data = {
    "key1": "value1",
    "key2": [1, 2, 3],
    "key3": {"nested": "object"}
}

with open(test_json_path, 'w') as f:
    json.dump(test_data, f)

ARGS = {
    "path": test_json_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
rotate_as_if_first Ran
ran; our contract draft was wrong, not the code
cstankonrad/long_llama/src/modeling_longllama.py
code served (permissive licence) · get_code("a13c903093d13c1f")
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) and returned Tensor, ndim 4, dtype family float, shape (2, 2, 3, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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 RotaryEmbedding(nn.Module):
    """Minimal rotary embedding that returns cos and sin tensors."""
    def __init__(self, dim):
        super().__init__()
        self.dim = dim
    
    def forward(self, x, seq_len):
        # x shape: (batch_size, seq_len, dim)
        # Return cos and sin with shapes that rotate_one expects
        # rotate_one does: cos.squeeze(1).squeeze(0) and sin.squeeze(1).squeeze(0)
        # Then indexes with position_ids of shape (batch_size, seq_len)
        # So cos/sin need to be indexable by position_ids
        
        # Create cos and sin tensors
        # After squeeze(1).squeeze(0), they should be 1D for indexing
        # So before squeezing, they should be (1, 1, seq_len, dim)
        cos = torch.ones(1, 1, seq_len, self.dim)
        sin = torch.ones(1, 1, seq_len, self.dim) * 0.5
        return cos, sin

# Create minimal rotary embedding
rotary_emb = RotaryEmbedding(dim=4)

# Create input tensor: (batch_size=2, seq_len=3, hidden_dim=4)
x = torch.randn(2, 3, 4)

ARGS = {
    "x": x,
    "rotary_emb": rotary_emb,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 2, 3, 4) — shape (2, 2, 3, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
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
rotate_half Ran
ran; our contract draft was wrong, not the code
CStanKonrad/long_llama/src/modeling_longllama.py
code served (permissive licence) · get_code("b99eea6376d1e212")
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) and returned Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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

torch.manual_seed(0)

# The function rotate_half takes a single parameter 'x' which is a tensor
# It accesses x.shape[-1] to get the last dimension
# It uses indexing with [..., :] slicing notation
# It calls torch.cat() to concatenate tensors

# Create a small tensor with shape (2, 4) - batch size 2, hidden dim 4
x = torch.randn(2, 4)

ARGS = {
    "x": x
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
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
rotate_one Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
cstankonrad/long_llama/src/modeling_longllama.py
code served (permissive licence) · get_code("feba6513abb9fb9c")
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) and returned Tensor, ndim 4, dtype family float, shape (2, 2, 3, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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

torch.manual_seed(0)

# Construct arguments for rotate_one function
# Parameters: x, cos, sin, position_ids

# x: tensor with shape [batch_size, seq_len, hidden_dim]
# The function uses x[..., :x.shape[-1] // 2] and x[..., x.shape[-1] // 2:]
# so hidden_dim must be even. Let's use batch=2, seq_len=3, hidden_dim=4
x = torch.randn(2, 3, 4)

# cos: tensor that will be squeezed and indexed by position_ids
# After squeeze(1).squeeze(0), it should have shape [max_position, hidden_dim]
# Then cos[position_ids] is called where position_ids has shape [batch_size, seq_len]
# So cos needs shape that after squeezing becomes [seq_len_or_larger, hidden_dim]
# Let's make it [1, 1, 5, 4] so after squeeze(1).squeeze(0) it's [5, 4]
cos = torch.randn(1, 1, 5, 4)

# sin: same shape as cos
sin = torch.randn(1, 1, 5, 4)

# position_ids: shape [batch_size, seq_len] with values in range [0, max_position)
# batch_size=2, seq_len=3, values should be < 5 (the first dimension of cos/sin after squeezing)
position_ids = torch.tensor([[0, 1, 2], [1, 2, 3]], dtype=torch.long)

ARGS = {
    "x": x,
    "cos": cos,
    "sin": sin,
    "position_ids": position_ids,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 2, 3, 4) — shape (2, 2, 3, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
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
apply_rotary_emb Not yet run
unverified (harvested, no recorded run)
CStanKonrad/long_llama/fot_continued_pretraining/EasyLM/models/llama/llama_model.py
code served (permissive licence) · get_code("a305ca669a9cacf3")
How it ran

no invocation recorded.

apply_rotary_emb_as_fst Not yet run
unverified (harvested, no recorded run)
CStanKonrad/long_llama/fot_continued_pretraining/EasyLM/models/llama/llama_model.py
code served (permissive licence) · get_code("fd8c57f88ddcfa62")
How it ran

no invocation recorded.

get_padding_token Not yet run
unverified (harvested, no recorded run)
CStanKonrad/long_llama/instruction_fine_tuning/data_processing.py
code served (permissive licence) · get_code("21a812da7fd6cdd2")
How it ran

no invocation recorded.

inverse_permute Not yet run
unverified (harvested, no recorded run)
CStanKonrad/long_llama/fot_continued_pretraining/EasyLM/models/llama/convert_hf_to_easylm.py
code served (permissive licence) · get_code("c2234099a81ca2e6")
How it ran

no invocation recorded.

tokenize_text_no_special_tokens Not yet run
unverified (harvested, no recorded run)
CStanKonrad/long_llama/instruction_fine_tuning/data_processing.py
code served (permissive licence) · get_code("630a45cda4fbc26b")
How it ran

no invocation recorded.

Also present in other papers' code

Of 14 samples lifted from this paper's linked repositories, 14 carry a function identity and 0 do not. Of the 14 distinct functions with an identity, 8 are also present in other papers whose code we hold (7 with at least one paper dated earlier, 2 the same month, 4 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.

FunctionAlso present inEarliest-datedOthers named
rotate_half 360 other papers
10 dated earlier, 5 the same month, 342 later, 3 undated · 30 only as an assembly brick
1901.02860 (2019-01)
Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context
2203.08913 (2022-03)
Memorizing Transformers
2302.13971 (2023-02)
LLaMA: Open and Efficient Foundation Language Models
and 357 more not named here
read_json 21 other papers
5 dated earlier, 1 the same month, 15 later
2210.07316 (2022-10)
MTEB: Massive Text Embedding Benchmark
2304.08177 (2023-04)
Efficient and Effective Text Encoding for Chinese LLaMA and Alpaca
2304.14454 (2023-04)
PMC-LLaMA: Towards Building Open-source Language Models for Medicine
and 18 more not named here
apply_rotary_emb 2 other papers
1 dated earlier, 1 later
2302.02676 (2023-02)
Chain of Hindsight Aligns Language Models with Feedback
2402.08268 (2024-02)
World Model on Million-Length Video And Language With Blockwise RingAttention
match_keywords 1 other paper
1 later
2406.09279 (2024-06)
Unpacking DPO and PPO: Disentangling Best Practices for Learning from Preference Feedback
mem_apply_update 1 other paper
1 dated earlier
2305.16300 (2023-05)
Random-Access Infinite Context Length for Transformers
precompute_freqs_cis 1 other paper
1 dated earlier
2302.02676 (2023-02)
Chain of Hindsight Aligns Language Models with Feedback
rotate_as_if_first 1 other paper
1 dated earlier
2305.16300 (2023-05)
Random-Access Infinite Context Length for Transformers
rotate_one 1 other paper
1 dated earlier
2305.16300 (2023-05)
Random-Access Infinite Context Length for Transformers

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

Large language models have an exceptional capability to incorporate new information in a contextual manner. However, the full potential of such an approach is often restrained due to a limitation in the effective context length. One solution to this issue is to endow an attention layer with access to an additional context, which comprises of (key, value) pairs. Yet, as the number of documents increases, the proportion of relevant keys to irrelevant ones decreases, leading the model to focus more on the irrelevant keys. We identify a significant challenge, dubbed the distraction issue, where keys linked to different semantic values might overlap, making them hard to distinguish. To tackle this problem, we introduce the Focused Transformer (FOT), a technique that employs a training process inspired by contrastive learning. This novel approach enhances the structure of the (key, value) space, enabling an extension of the context length. Our method allows for fine-tuning pre-existing, large-scale models to lengthen their effective context. This is demonstrated by our fine-tuning of 3B and 7B OpenLLaMA checkpoints. The resulting models, which we name LONGLLAMA 2 , exhibit advancements in tasks requiring a long context. We further illustrate that our LONGLLAMA models adeptly manage a 256k context length for passkey retrieval. * Equal contribution † Work done while at Google Research. 2 We release the checkpoints and source code of LONGLLAMA , see also our colabs. 37th Conference on Neural Information Processing Systems (NeurIPS 2023).

For agents

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

get_harvested_code_for_paper("2307.03170")
get_code_for_paper("2307.03170")
have("2307.03170")

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

curl https://syntology.ai/api/ran/2307.03170.json

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

[![Syntology run record](https://syntology.ai/api/ran/2307.03170.svg)](https://syntology.ai/paper/2307.03170)

Connect an agent — have() is free.