Pith. sign in

REVIEW 5 major objections 5 minor 30 references

The Impact of Model Scaling on Seen and Unseen Language Performance

T0 review · 5 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Multilingual scaling laws are not universal: across 204 languages, model size has little effect on zero-shot text classification, produces linear gains in two-shot classification, and helps translation only for an instruction-tuned model.

desk verdict Broad 204-language scaling study with a useful classification map, but the translation finding for BLOOMZ is confounded by training/eval overlap and the abstract overstates the resource-level result. read the letter →

arxiv 2501.05629 v1 pith:VHDUI5V7 submitted 2025-01-10 cs.CL cs.AI

Rhitabrat Pokharel , Sina Bagheri Nezhad , Ameeta Agrawal , Suresh Singh This is my paper · ORCID
classification cs.CLcs.AI
keywords multilinguallanguagemodelsmodelscalingzero-shotevaluationfew-shotin-contextlearninglow-resourcelanguagestextclassificationmachinetranslationseenandunseen
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper sets out to answer a practical question: when a multilingual language model is made larger, do languages see a performance improvement, and does it matter whether the language appeared in pretraining? The authors evaluate three model families (XGLM, BLOOM, BLOOMZ) at parameter counts from 560 million to 7.5 billion across 204 languages on topic classification and machine translation, in zero-shot and two-shot settings. They find that scaling does not follow one trend: zero-shot classification stays mostly flat as models grow, two-shot classification shows clear linear gains with scale, and translation improves with scale only for the instruction-tuned BLOOMZ. Seen languages outperform unseen languages throughout, and a language's general resource level predicts performance better than its share of pretraining data. If this is right, developers cannot assume a universal scaling law for multilingual systems; the task, the prompting setup, and pretraining visibility determine whether scale helps.

What carries the argument

The argument is carried by the empirical scaling curve: each model family comes in several sizes trained on the same corpus, so task performance can be plotted against parameter count and summarized by the slope of a linear fit. The comparisons are organized by a seen-versus-unseen split built from pretraining language lists and by six resource levels, applied to both tasks under zero-shot and two-shot prompting. The slopes in Table 3 — near zero for zero-shot classification, positive for two-shot classification, and near zero for most translation settings — are the evidence for the central claim.

What would settle it

Obtain the actual pretraining corpora for XGLM, BLOOM, and BLOOMZ and check whether the languages labeled 'potentially unseen' actually appear in them; if a substantial fraction do, the flat zero-shot scaling and the seen-unseen gaps are artifacts of the labeling. Recomputing the scaling slopes using only languages with unambiguous ISO and script mappings would give a direct test.

Watch

Extended reading notes

Core claim

The paper's central claim is that multilingual scaling behavior is task- and setting-dependent rather than universal. In text classification on SIB-200, macro-F1 is largely flat from 560M to 7.5B parameters under zero-shot prompting; under two-shot prompting, larger models show clear linear improvement for both seen and unseen languages, with XGLM's unseen-language performance as the one slight exception. In xx-to-English translation on FLORES-200, XGLM and BLOOM show little or no scaling, whereas the instruction-tuned BLOOMZ shows clear improvements from scale under both settings, although two-shot demonstrations hurt it. Seen languages consistently beat potentially unseen languages, including within the same language family, and larger models narrow the gap for seen languages but not for unseen ones. The paper further claims that overall resource level, rather than the language's proportion in pretraining data, is the stronger predictor of performance, and that this is true for seen languages but not for unseen ones.

Load-bearing premise

The load-bearing premise is that the pretraining language lists used to label each language as seen or unseen are accurate, and that the ISO-code and script mapping that connects dataset languages to those lists is correct; if a language is miscategorized, every seen-versus-unseen comparison and the scaling conclusions built on it become unreliable.

Editorial extensions

If this is right

  • Zero-shot multilingual classification benchmarks will show little separation across model sizes from 560M to 7.5B; users should expect flat scores without demonstrations.
  • Two-shot demonstrations are what unlock scaling benefits in classification, so few-shot evaluation is necessary to observe improvements from larger models.
  • For translation, increasing model size alone will not reliably improve low-resource quality; instruction tuning is the observed exception.
  • Resource-level groupings, not pretraining corpus proportions, should guide expectations for a language's performance.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Inference: if overall resource level is the stronger predictor, then adding pretraining data for a single low-resource language may not move its score much; improving related-language coverage or general data diversity could matter more.
  • Inference: the k=2 ceiling may itself shape the result; testing larger k on models with longer context windows could reveal whether translation scaling emerges later.
  • Inference: the contrast between flat zero-shot and linear two-shot classification suggests the scaling signal lies in in-context learning capacity rather than static multilingual knowledge; comparing English-prompt versus target-language-prompt results would test this.
  • Inference: because BLOOMZ translation degrades when given two examples, instruction-tuned models may be sensitive to prompt format; varying example ordering or templates could restore the benefit.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper studies scaling behavior of three multilingual model families (XGLM, BLOOM, BLOOMZ) across 204 languages on two tasks: topic classification with SIB-200 and machine translation with FLORES-200. Models are evaluated at multiple sizes under zero-shot and two-shot prompting, and languages are categorized as seen or unseen based on the models' reported pretraining language lists. The paper reports scaling slopes, resource-level correlations, and claims that scaling effects depend strongly on task and setting: zero-shot performance is mostly flat, two-shot classification improves linearly with size, and among translation models only the instruction-tuned BLOOMZ shows clear scaling benefits. The resource-level analysis claims that general resource levels predict performance better than language-specific pretraining data proportions.

Significance. If the findings are robust, the paper would provide a useful large-scale empirical map of how model size, pretraining visibility, and prompting interact across more than 200 languages, extending prior scaling studies that cover far fewer languages. The use of publicly listed pretraining language distributions, open model families, and two standardized multilingual benchmarks is a strength, as is the breadth of the evaluation (over two million scored instances). However, the central claims currently rest on several load-bearing assumptions and statistical procedures that are not yet supported: a potential train/evaluation overlap for BLOOMZ on FLORES-200, a fragile seen/unseen mapping, and slope and correlation estimates computed from very few model sizes without uncertainty quantification. These issues materially affect the headline conclusions rather than being presentation concerns.

major comments (5)
  1. [Seen and Unseen Languages] The abstract's translation claim ('only the instruction-tuned model showed clear benefits from scaling') rests on the BLOOMZ results, but the paper itself states in the Text generation section that part of the FLORES-200 dataset was used to instruction-tune BLOOMZ. Because the translation evaluation uses FLORES-200, the observed scaling slope could be inflated by memorization of instruction-tuning examples if any of the 204 test sentences per language overlap with xP3 training data. The manuscript never quantifies this overlap nor reports results on a verified disjoint subset. Please either demonstrate that the evaluation instances are disjoint from the instruction-tuning data, or re-estimate the BLOOMZ translation scaling on a clean subset and state whether the 'clear benefits' conclusion survives.
  2. [Table 3; Figure 1] The seen/unseen dichotomy is load-bearing for most comparisons, but it depends entirely on the accuracy of the model-card pretraining language lists and on an ISO 639-3 to ISO 639-1 mapping that resolves multiple scripts using 'the most common script type.' The paper itself qualifies these languages as 'potentially unseen.' Any miscategorization would change every seen-versus-unseen comparison, including the main scaling disparities. Please provide a per-language mapping table, justify the script-resolution decisions, and report a sensitivity analysis that excludes ambiguous languages or uses stricter criteria for 'unseen' status.
  3. [Appendix; Table 4] The central scaling statements are based on linear fits to only four or five model sizes per family and setting, yet Table 3 reports slopes without confidence intervals, standard errors, or significance tests. In particular, the claims that zero-shot performance is 'mostly flat' and that two-shot classification shows 'clear linear improvements' require interval estimates to distinguish genuine trends from noise around a small number of points. Please add uncertainty quantification (for example, bootstrap intervals over languages or model sizes) and, where possible, permutation or correlation tests to support the qualitative claims.
  4. [Prompts, Zero-shot, and Few-shot In-context Learning] The abstract's claim that 'overall resource levels, not just the proportions of pretraining languages, are better predictors of model performance' is supported only by Pearson correlations in Table 4, with no confidence intervals, p-values, or comparisons between dependent correlations. The table also aggregates over a small number of seen languages (30 for XGLM, roughly 45 for BLOOM/BLOOMZ), and the manuscript notes that for unseen languages the correlations are poor. Please report significance tests and interval estimates for the PD, RL, and RL* correlations, and frame the 'better predictor' claim according to what those tests support.
  5. Each language contributes only 204 test sentences, and few-shot demonstrations are drawn from a single random sample. Per-language F1 and especially per-language SacreBLEU scores computed from 204 sentences are noisy, yet the analysis uses them as fixed points when fitting slopes and resource-level correlations. Please assess stability with respect to the demonstration sample (for example, by repeating with multiple random seeds) or with per-language bootstrap intervals, and state how the conclusions change if noisier low-resource languages are downweighted.
minor comments (5)
  1. [Conclusion] The sentence 'In Section we saw that few-shot ICL hurts smaller models' has an incomplete section reference; please supply the actual section number or name.
  2. [Figure 4 caption] The caption renders '& - seen; & - unseen' instead of a readable legend convention; please define the line styles explicitly in the caption text.
  3. [Appendix; Table 4] The text says 'Similar correlation were obtained for 2-shot' but no corresponding table or numbers are provided; either include the 2-shot correlations or remove the unsupported statement.
  4. [Text generation using bloomz; Table 3] Table 3 omits BLOOMZ rows for text generation because BLOOMZ is plotted separately, but the abstract's translation claim depends on that excluded model; consider adding the BLOOMZ generation slopes to the table with an explicit contamination caveat.
  5. [Prompts, Zero-shot, and Few-shot In-context Learning] The prompt templates used for classification and translation are referenced only by prior work; including the exact prompt strings in an appendix would improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper is an observational benchmark study whose fits are descriptive, not predictions, and the acknowledged BLOOMZ/FLORES-200 overlap is a validity confound rather than a circular step.

full rationale

This paper is an empirical evaluation study, not a derivation. It reports measurements of F1 and SacreBLEU across model families, sizes, languages, and shot settings, and it fits linear slopes only to summarize the observed scaling trends (Table 3). These slopes are descriptive summaries of the same data, not predictions of held-out quantities, so no fitted-input-called-prediction step is present. The seen/unseen categorization is an input assumption based on model-card language lists, and the resource-level correlation analysis is in-sample, but neither makes the central claims definitionally equal to their inputs. The paper explicitly flags one genuine validity concern: part of the FLORES-200 dataset overlaps with BLOOMZ instruction-tuning data, stated as 'We do not show the results of bloomz in this plot and instead plot them separately later because part of the Flores-200 dataset was used to instruction-tune bloomz.' That overlap could confound the translation-scaling claim for BLOOMZ, but it is a data-contamination and external-validity issue, not a circular construction: the observed scaling behavior is not equivalent by definition to the evaluation data or to any parameter fitted from it. The paper also contains no load-bearing self-citation chain, no imported uniqueness theorem, and no ansatz smuggled in via citation. The central findings are external measurements with stated assumptions, so the appropriate verdict is no significant circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The main empirical findings rest on benchmark reliability, an accurate seen/unseen mapping, English prompts, and within-family comparability. The quantitative claims also depend on fitted slopes and hand-chosen resource-level thresholds that are reported without uncertainty. No new entities are postulated.

free parameters (3)
  • Linear scaling slope per condition = Values in Table 3, e.g., 0.0718 for bloomz 2-shot seen classification
    The central scaling claims are based on slopes of linear fits over four to five model sizes; these slopes are fitted to the data and reported without confidence intervals.
  • Resource-level merge threshold = low = levels 0-2, high = levels 3-5
    Used to compute the merged resource-level correlations RL* in Table 4; the split is chosen by hand and not justified.
  • Number of demonstrations k = 2
    The two-shot setting is the main few-shot condition; the value was chosen because longer prompts exceed the 2048-token context window of some models, and the paper acknowledges exploring more shots was out of scope.
assumptions (4)
  • domain assumption SIB-200 and FLORES-200 benchmarks provide reliable language-level measurements with 204 test sentences per language.
    All F1 and BLEU numbers are computed from these 204-instance test sets; the paper does not validate stability of the estimates, and only one demonstration sample is used.
  • domain assumption Model-card pretraining language lists and the ISO mapping correctly identify seen languages.
    The paper manually maps xglm ISO 639-1 languages to dataset ISO 639-3 codes and picks the most common script for ambiguous cases; errors would change the seen/unseen split.
  • domain assumption English prompts are appropriate for all 204 target languages and do not distort scaling comparisons.
    The paper cites prior work for English prompts but does not validate prompt language per language; prompt effectiveness could interact with model scale.
  • domain assumption Model sizes within each family are comparable because they share pretraining data and architecture.
    The paper states this for xglm and bloom/bloomz, but bloomz additionally underwent instruction tuning, which the paper treats as a separate family.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The Impact of Model Scaling on Seen and Unseen Language Performance." pith.science (2026). https://pith.science/paper/VHDUI5V7

@misc{pith2026250105629,
  author       = {Pith},
  title        = {Pith review of: The Impact of Model Scaling on Seen and Unseen Language Performance},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VHDUI5V7}},
  note         = {Machine review of arXiv:2501.05629}
}
read the original abstract

The rapid advancement of Large Language Models (LLMs), particularly those trained on multilingual corpora, has intensified the need for a deeper understanding of their performance across a diverse range of languages and model sizes. Our research addresses this critical need by studying the performance and scaling behavior of multilingual LLMs in text classification and machine translation tasks across 204 languages. We systematically examine both seen and unseen languages across three model families of varying sizes in zero-shot and few-shot settings. Our findings show significant differences in scaling behavior between zero-shot and two-shot scenarios, with striking disparities in performance between seen and unseen languages. Model scale has little effect on zero-shot performance, which remains mostly flat. However, in two-shot settings, larger models show clear linear improvements in multilingual text classification. For translation tasks, however, only the instruction-tuned model showed clear benefits from scaling. Our analysis also suggests that overall resource levels, not just the proportions of pretraining languages, are better predictors of model performance, shedding light on what drives multilingual LLM effectiveness.

Figures

Figures reproduced from arXiv: 2501.05629 by the authors.

Figure 1
Figure 1. Multilingual performance under 0-shot and 2-shot settings. Figure 1a shows results for text classification using SIB-200 [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Performance of three models ( xglm; bloom; bloomz) on English-only subset of SIB-200. On the x-axis are the model sizes [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. bloomz’s performance on seen and unseen lan￾guages for text generation task. ICL brings additional improvements for all models except for xglm unseen. Nonetheless, these results suggest that mul￾tilingual scaling trends for text classification task may differ depending on the settings adopted (0- vs. 2-shot). This is similar to our analysis of baseline English-only evaluation behavior (as shown in [PITH_FULL_IMAGE:… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: The top row shows performance comparison between seen and unseen for languages on the [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Classification task: Results of evaluation (F1 score) across different models based on language resource level using SIB-200 dataset. The model sizes on the x-axis are in billions of parameters. Resource Level 0, Resource Level 1, Resource Level 2, Resource Level 3, Re…
Figure 6
Figure 6. Figure 6: Generation task: Results of evaluation (SacreBLEU) across different models based on language resource level using Flores-200 dataset. The model sizes on the x-axis are in billions of parameters. Resource Level 0, Resource Level 1, Resource Level 2, Resource Level 3, Re…
Figure 7
Figure 7. Figure 7: Relative improvement in log-scale on y-axis from [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Results of evaluation (F1 score) across different models based on language resource level for the [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: Results of evaluation (SacreBLEU) across different models based on language resource level for the [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 11 canonical work pages

  1. [1]

    This is reflected in how these models perform in the languages of the different resource levels

    The strongest correlation between the performance of seen languages and their resource levels is shown by bloom, followed bloomz, and lastly, xglm. This is reflected in how these models perform in the languages of the different resource levels. Our discussion carries forth to the generation task as well, with one notable exception. We see that xglm for th...

    work page 2022

  2. [3]

    In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing

    MEGA: Multilingual Evaluation of Generative AI. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. Singa- pore: Association for Computational Linguistics. Anil, R.; Dai, A. M.; Firat, O.; Johnson, M.; Lepikhin, D.; Passos, A.; et al

    work page 2023

  3. [4]

    arXiv:2305.10403

    PaLM 2 Technical Report. arXiv:2305.10403. Asai, A.; Kudugunta, S.; Yu, X. V .; Blevins, T.; Gonen, H.; Reid, M.; et al

    arXiv

  4. [5]

    arXiv preprint arXiv:2305.14857

    Buffet: Benchmarking large language models for few-shot cross-lingual transfer. arXiv preprint arXiv:2305.14857. Bach, S.; Sanh, V .; Yong, Z. X.; Webson, A.; Raffel, C.; Nayak, N. V .; et al

    arXiv

  5. [6]

    English Prompts are Better for NLI-based Zero-Shot Emotion Classification than Target-Language Prompts

    English Prompts are Better for NLI-based Zero-Shot Emotion Clas- sification than Target-Language Prompts. arXiv preprint arXiv:2402.03223. Brown, T.; Mann, B.; Ryder, N.; Subbiah, M.; Kaplan, J. D.; Dhariwal, P.; et al. 2020a. Language Models are Few-Shot Learners. In Advances in Neural Information Processing Systems, volume 33, 1877–1901. Curran Associat...

    work page Pith review arXiv 1901

  6. [7]

    In Miceli-Barone, A

    In- structEval: Towards Holistic Evaluation of Instruction-Tuned Large Language Models. In Miceli-Barone, A. V .; Barez, F.; Cohen, S.; V oita, E.; Germann, U.; and Lukasik, M., eds., Proceedings of the First edition of the Workshop on the Scal- ing Behavior of Large Language Models (SCALE-LLM 2024), 35–64. St. Julian’s, Malta: Association for Computation...

    work page 2024

  7. [9]

    Understanding BLOOM: An empirical study on diverse NLP tasks

    Under- standing BLOOM: An empirical study on diverse NLP tasks. arXiv preprint arXiv:2211.14865. Etxaniz, J.; Azkune, G.; Soroa, A.; de Lacalle, O. L.; and Artetxe, M

    work page Pith review arXiv

  8. [10]

    Fernandes, P.; Ghorbani, B.; Garcia, X.; Freitag, M.; and Fi- rat, O

    Do Multilingual Language Models Think Better in English? arXiv preprint arXiv:2308.01223. Fernandes, P.; Ghorbani, B.; Garcia, X.; Freitag, M.; and Fi- rat, O

    arXiv

Show all 30 references
  1. [12]

    arXiv:2203.15556

    Training Compute- Optimal Large Language Models. arXiv:2203.15556. ImaniGooghari, A.; Lin, P.; Kargaran, A. H.; Severini, S.; Jalili Sabet, M.; Kassner, N.; et al

    arXiv

  2. [13]

    arXiv:2402.04177

    Scaling Laws for Downstream Task Performance of Large Language Models. arXiv:2402.04177. Joshi, P.; Santy, S.; Budhiraja, A.; Bali, K.; and Choudhury, M

  3. [15]

    In Findings of the Association for Computational Linguistics: EMNLP 2023

    ChatGPT Beyond English: Towards a Comprehensive Evaluation of Large Lan- guage Models in Multilingual Learning. In Findings of the Association for Computational Linguistics: EMNLP 2023 . Singapore: Association for Computational Linguistics. Lin, X. V .; Mihaylov, T.; Artetxe, ...

    2023

  4. [16]

    In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Pro- cessing

    Few-shot Learning with Multilingual Generative Language Models. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Pro- cessing. Abu Dhabi, United Arab Emirates: Association for Computational Linguistics. McKenzie, I. R.; Lyzhov, A.; Pieler, M. M.; ...

    2022

  5. [17]

    In-context Learn- ing: A Fair Comparison and Evaluation

    Few-shot Fine-tuning vs. In-context Learn- ing: A Fair Comparison and Evaluation. arXiv:2305.16938. Muennighoff, N.; Wang, T.; Sutawika, L.; Roberts, A.; Bider- man, S.; Le Scao, T.; et al

    arXiv

  6. [19]

    arXiv preprint arXiv:2207.04672

    No Language Left Be- hind: Scaling Human-Centered Machine Translation. arXiv preprint arXiv:2207.04672. Pfeiffer, J.; Goyal, N.; Lin, X.; Li, X.; Cross, J.; Riedel, S.; and Artetxe, M

    arXiv

  7. [20]

    In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Lan- guage Technologies

    Lifting the Curse of Multilingual- ity by Pre-training Modular Transformers. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Lan- guage Technologies. Seattle, United States: Association for Computation...

    2022

  8. [21]

    arXiv:2112.11446

    Scaling Language Mod- els: Methods, Analysis & Insights from Training Gopher. arXiv:2112.11446. Sampathkumar, R.; Kravitz, E.; and Huang, X

    arXiv

  9. [22]

    arXiv:2211.05100

    BLOOM: A 176B-Parameter Open-Access Multilingual Language Model. arXiv:2211.05100. Shliazhko, O.; Fenogenova, A.; Tikhonova, M.; Kozlova, A.; Mikhailov, V .; and Shavrina, T

    arXiv

  10. [23]

    arXiv preprint arXiv:2201.11990

    Using deepspeed and mega- tron to train megatron-turing nlg 530b, a large-scale genera- tive language model. arXiv preprint arXiv:2201.11990. Srivastava, A.; Rastogi, A.; Rao, A.; Shoeb, A. A. M.; Abid, A.; Fisch, A.; et al

    arXiv

  11. [24]

    In Miceli-Barone, A

    Scaling Behavior of Machine Translation with Large Language Models under Prompt Injection Attacks. In Miceli-Barone, A. V .; Barez, F.; Cohen, S.; V oita, E.; Germann, U.; and Lukasik, M., eds., Proceedings of the First edition of the Workshop on the Scal- ing Behavior of Larg...

    2024

  12. [25]

    arXiv:2302.13971

    LLaMA: Open and Efficient Foundation Language Models. arXiv:2302.13971. Wei, J.; Kim, N.; Tay, Y .; and Le, Q

    arXiv

  13. [26]

    In Proceedings of the 2023 Confer- ence on Empirical Methods in Natural Language Processing

    Inverse Scaling Can Become U-Shaped. In Proceedings of the 2023 Confer- ence on Empirical Methods in Natural Language Processing. Singapore: Association for Computational Linguistics. Wei, J.; Tay, Y .; Bommasani, R.; Raffel, C.; Zoph, B.; Borgeaud, S.; et al

    2023

  14. [27]

    arXiv:2206.07682

    Emergent Abilities of Large Lan- guage Models. arXiv:2206.07682. Winata, G.; Wu, S.; Kulkarni, M.; Solorio, T.; and Preot ¸iuc- Pietro, D

    arXiv

  15. [28]

    arXiv preprint arXiv:2212.09535

    Bloom+ 1: Adding language support to bloom for zero-shot prompting. arXiv preprint arXiv:2212.09535. Zhang, Y .; Yasunaga, M.; Zhou, Z.; HaoChen, J. Z.; Zou, J.; Liang, P.; and Yeung, S

    arXiv

  16. [29]

    In Rogers, A.; Boyd-Graber, J.; and Okazaki, N., eds., Findings of the Association for Computational Linguistics: ACL 2023, 7479–7498

    Beyond Positive Scaling: How Negation Impacts Scaling Trends of Language Models. In Rogers, A.; Boyd-Graber, J.; and Okazaki, N., eds., Findings of the Association for Computational Linguistics: ACL 2023, 7479–7498. Toronto, Canada: Association for Computational Linguistics. A...

    2020

  17. [2017]

    arXiv:1712.00409

    Deep Learning Scaling is Pre- dictable, Empirically. arXiv:1712.00409. Hoffmann, J.; Borgeaud, S.; Mensch, A.; Buchatskaya, E.; Cai, T.; Rutherford, E.; et al

    arXiv

  18. [2019]

    arXiv:1912.02292

    Deep Double Descent: Where Bigger Models and More Data Hurt. arXiv:1912.02292. NLLB Team; Costa-juss`a, M. R.; Cross, J.; C ¸elebi, O.; El- bayad, M.; Heafield, K.; et al

    1912 arXiv

  19. [2020]

    arXiv:2001.08361

    Scaling Laws for Neu- ral Language Models. arXiv:2001.08361. Lai, V .; Ngo, N.; Pouran Ben Veyseh, A.; Man, H.; Dernon- court, F.; Bui, T.; and Nguyen, T

    2001 arXiv

  20. [2022]

    arXiv:2204.02311

    PaLM: Scaling Language Mod- eling with Pathways. arXiv:2204.02311. Conneau, A.; Khandelwal, K.; Goyal, N.; Chaudhary, V .; Wenzek, G.; Guzm´an, F.; et al

    arXiv

  21. [2023]

    arXiv:2303.08774

    GPT-4 Technical Report. arXiv:2303.08774. Adelani, D. I.; Liu, H.; Shen, X.; Vassilyev, N.; Alabi, J. O.; Mao, Y .; et al

    arXiv

  22. [2024]

    arXiv:2309.07445

    SIB-200: A Simple, Inclusive, and Big Evaluation Dataset for Topic Classification in 200+ Lan- guages and Dialects. arXiv:2309.07445. Ahuja, K.; Diddee, H.; Hada, R.; Ochieng, M.; Ramesh, K.; Jain, P.; et al

    arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.