Modelpublic

aitf-komdigi/KomdigiITS-8B-DFK-MultimodalClassification

sourceHugging Faceupdated 4mo agoView on Hugging Face

0likes

Model Card

font-family: var(--sans); color: var(--text); max-width: 880px; margin: 0 auto; padding: 0 0 64px; line-height: 1.7; font-size: 1rem; background: radial-gradient(ellipse at 50% 0%, rgba(212,160,74,0.03) 0%, transparent 50%), var(--bg); }

/ ── Hero ── / .df-hero { position: relative; overflow: hidden; } .df-hero img { display: block; width: 100%; height: 340px; object-fit: cover; margin: 0; } .df-ident { position: absolute; bottom: 0; left: 0; right: 0; padding: 100px 36px 32px; background: linear-gradient(to top, var(--bg) 0%, rgba(12,18,32,0.92) 40%, transparent 100%); } .df-name { font-size: 2.6rem; font-weight: 900; color: var(--bright); letter-spacing: 0.04em; line-height: 1.05; margin: 0 0 10px; text-shadow: 0 2px 8px rgba(0,0,0,0.5); overflow-wrap: break-word; } .df-base { font-family: var(--mono); font-size: 0.66rem; color: var(--accent); letter-spacing: 0.14em; text-transform: uppercase; display: block; }

/ ── Sections ── / .df-section { padding: 0; } .df-shead { position: relative; display: flex; align-items: center; gap: 14px; padding: 16px 36px 14px; margin-bottom: 24px; border-top: 2px solid; border-image: linear-gradient(90deg, var(--accent), var(--warn)) 1; } .df-snum { font-family: var(--mono); font-size: 2rem; font-weight: 900; color: var(--accent); letter-spacing: 0.06em; opacity: 0.10; position: absolute; right: 36px; top: 50%; transform: translateY(-50%); line-height: 1; } .df-stitle { font-size: 1.05rem; font-weight: 700; letter-spacing: 0.1em; text-transform: uppercase; color: var(--bright); } .df-stitle::before { content: '\2726'; color: var(--accent); font-size: 0.8em; margin-right: 8px; } .df-sbody { padding: 0 36px 40px; } .df-sbody p { margin: 0 0 14px; font-size: 0.94rem; } .df-sbody p:last-child { margin-bottom: 0; }

/ ── Sub-headings ── / .df-sub { color: var(--bright) !important; font-size: 0.95rem !important; margin: 22px 0 12px !important; padding: 0 0 7px !important; font-weight: 700; text-transform: uppercase; letter-spacing: 2px; border: none !important; border-bottom: 1px solid var(--edge) !important; }

/ ── Panels ── / .df-grid { display: grid; grid-template-columns: 1fr 1fr; gap: 16px; } .df-panel { border: 1px solid var(--edge); border-left: 3px solid var(--accent); position: relative; background: var(--surface); box-shadow: 0 0 16px rgba(212,160,74,0.03); } .df-panel::before { content: ''; position: absolute; top: -1px; right: -1px; width: 9px; height: 9px; border-top: 1px solid var(--accent); border-right: 1px solid var(--accent); opacity: 0.45; } .df-panel::after { content: ''; position: absolute; bottom: -1px; right: -1px; width: 9px; height: 9px; border-bottom: 1px solid var(--warn); border-right: 1px solid var(--warn); opacity: 0.35; } .df-panel--warn { border-left-color: var(--warn); } .df-phead { font-family: var(--mono); font-size: 0.66rem; font-weight: 700; letter-spacing: 0.14em; text-transform: uppercase; color: var(--dim); padding: 9px 14px; border-bottom: 1px solid var(--edge); } .df-phead::after { content: ' \2726'; color: var(--accent); opacity: 0.4; } .df-row { display: grid; grid-template-columns: 13ch 1fr; align-items: baseline; column-gap: 4px; padding: 8px 14px; border-bottom: 1px solid var(--edge); font-size: 0.88rem; } .df-row:last-child { border-bottom: none; } .df-k { font-family: var(--mono); font-size: 0.88rem; color: var(--dim); } .df-k::after { content: ':'; } .df-v { color: var(--bright); font-size: 0.88rem; } .df-row .df-v:only-child { grid-column: 1 / -1; }

/ ── Metric cards ── / .df-metrics { display: grid; grid-template-columns: repeat(4, 1fr); gap: 10px; margin-bottom: 18px; } .df-metric { background: var(--surface); border: 1px solid var(--edge); border-top: 3px solid var(--accent); padding: 14px 10px 12px; text-align: center; box-shadow: 0 0 12px var(--ac-glow); position: relative; } .df-metric::before { content: ''; position: absolute; top: -1px; right: -1px; width: 8px; height: 8px; border-top: 1px solid var(--accent); border-right: 1px solid var(--accent); opacity: 0.3; } .df-metric--hi { border-top-color: var(--warn); } .df-metric--hi::before { border-color: var(--warn); } .df-mval { font-family: var(--mono); font-size: 1.6rem; font-weight: 900; color: var(--accent); line-height: 1; margin-bottom: 4px; } .df-metric--hi .df-mval { color: var(--warn); } .df-mlbl { font-family: var(--mono); font-size: 0.54rem; font-weight: 700; letter-spacing: 0.14em; text-transform: uppercase; color: var(--dim); }

/ ── Note box ── / .df-note { border: 1px solid var(--edge); border-left: 3px solid var(--accent); padding: 12px 16px; margin-top: 16px; background: var(--ac-glow); font-size: 0.86rem; color: var(--dim); position: relative; } .df-note::before { content: ''; position: absolute; top: -1px; right: -1px; width: 8px; height: 8px; border-top: 1px solid var(--accent); border-right: 1px solid var(--accent); opacity: 0.3; } .df-note strong { color: var(--accent); font-family: var(--mono); font-size: 0.68rem; letter-spacing: 0.1em; text-transform: uppercase; }

/ ── Links ── / .df a { color: var(--bright); text-decoration: none; border-bottom: 1px solid var(--rule); } .df a:hover { color: var(--accent); border-bottom-color: var(--accent); }

/ ── Dropdown ── / .df details { border: 1px solid var(--edge); border-left: 3px solid var(--accent); margin-top: 20px; position: relative; background: var(--surface); box-shadow: 0 0 16px rgba(212,160,74,0.03); } .df details::before { content: ''; position: absolute; top: -1px; right: -1px; width: 9px; height: 9px; border-top: 1px solid var(--accent); border-right: 1px solid var(--accent); opacity: 0.45; } .df details::after { content: ''; position: absolute; bottom: -1px; right: -1px; width: 9px; height: 9px; border-bottom: 1px solid var(--warn); border-right: 1px solid var(--warn); opacity: 0.35; } .df summary { list-style: none; padding: 10px 14px; cursor: pointer; font-family: var(--mono); font-size: 0.7rem; font-weight: 700; letter-spacing: 0.12em; text-transform: uppercase; color: var(--dim); user-select: none; display: flex; align-items: center; gap: 10px; } .df summary::-webkit-details-marker { display: none; } .df summary::before { content: '+'; color: var(--accent); font-size: 1rem; line-height: 1; flex-shrink: 0; } .df details[open] summary::before { content: '\2212'; } .df summary:hover { color: var(--bright); } .df-drop-body { padding: 18px 16px; border-top: 1px solid var(--edge); } .df-drop-body p { margin: 0 0 14px; font-size: 0.9rem; }

/ ── Code ── / .df pre { background: #080e18; border: 1px solid var(--edge); border-left: 2px solid var(--accent); padding: 14px 16px; overflow-x: auto; font-family: var(--mono); font-size: 0.74rem; line-height: 1.6; color: var(--text); margin: 0 0 18px; } .df pre:last-child { margin-bottom: 0; } .df pre code { background: none; color: inherit; padding: 0; border: none; } .df code { font-family: var(--mono); font-size: 0.85em; color: var(--accent); background: var(--ac-glow); padding: 2px 5px; border: 1px solid rgba(212,160,74,0.12); }

/ ── Mobile ── / @media (max-width: 640px) { .df-hero img { height: 220px; } .df-name { font-size: 1.8rem; } .df-ident { padding: 80px 20px 24px; } .df-shead { padding: 14px 20px 12px; } .df-sbody { padding: 0 20px 32px; } .df-snum { right: 20px; } .df-grid { grid-template-columns: 1fr; } .df-metrics { grid-template-columns: repeat(2, 1fr); } .df-row { grid-template-columns: 1fr; gap: 2px; } .df-k::after { content: ''; } } </style> <html lang="en"> <head> <meta charset="UTF-8"> <meta name="viewport" content="width=device-width, initial-scale=1.0"> <title>KomdigiITS-8B-DFK-MultimodalClassification</title> <link rel="preconnect" href="https://fonts.googleapis.com"> <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin> <link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;600;700;800;900&family=JetBrains+Mono:wght@400;500;700&display=swap" rel="stylesheet"> </head> <body> <div class="df">

<div class="df-hero"> <img src="dfkherobanner.png" alt="image"> <div class="df-ident"> <h1 class="df-name">KomdigiITS-8B-DFK<br>Multimodal Classification</h1> <span class="df-base">Ministral-3-8B-Base-2512 · LoRA · Vision-Language</span> </div> </div>

<div class="df-section"> <div class="df-shead"> <span class="df-snum">01</span> <span class="df-stitle">Overview</span> </div> <div class="df-sbody"> <p>A LoRA adapter fine-tuned on <a href="https://huggingface.co/aitf-komdigi/KomdigiITS-8B-DFK-CPT">aitf-komdigi/KomdigiITS-8B-DFK-CPT</a> (Ministral-3-8B-Base-2512 based) as a Vision-Language Model for multimodal content classification. The model analyzes social media screenshots and classifies them into four categories: <code>netral</code>, <code>disinformasi</code>, <code>fitnah</code>, and <code>ujaran kebencian</code>.</p> <p>Trained using the <a href="https://github.com/aitf-its-tim3-dfk/SITA">SITA</a> framework with Unsloth's SFT pipeline. Given an image, the model produces a structured analysis with a classification label and a detailed Indonesian-language reasoning of any violations found.</p> <div class="df-note"> <strong>♦ Note:</strong> This is the final checkpoint from Workshop 3 (<code>final-ministral-8b-cpt-ws3</code>), trained on the DFK VLM Dataset V3 with augmented train/val splits. The base model (<code>aitf-komdigi/KomdigiITS-8B-DFK-CPT</code>) was continual-pretrained on DFK domain-oriented text before fine-tuning. </div> </div> </div>

<div class="df-section"> <div class="df-shead"> <span class="df-snum">02</span> <span class="df-stitle">Model Details</span> </div> <div class="df-sbody"> <div class="df-grid"> <div class="df-panel"> <div class="df-phead">Identity</div> <div class="df-row"><span class="df-k">Developed</span><span class="df-v">DFK Tim 3 ITS</span></div> <div class="df-row"><span class="df-k">Type</span><span class="df-v">VLM — LoRA adapter</span></div> <div class="df-row"><span class="df-k">Language</span><span class="df-v">Indonesian</span></div> </div> <div class="df-panel"> <div class="df-phead">Architecture</div> <div class="df-row"><span class="df-k">Base</span><span class="df-v"><a href="https://huggingface.co/aitf-komdigi/KomdigiITS-8B-DFK-CPT">KomdigiITS-8B-DFK-CPT</a></span></div> <div class="df-row"><span class="df-k">Arch</span><span class="df-v">Mistral3ForConditionalGeneration</span></div> <div class="df-row"><span class="df-k">Params</span><span class="df-v">8B (base)</span></div> <div class="df-row"><span class="df-k">Precision</span><span class="df-v">float16</span></div> </div> </div> </div> </div>

<div class="df-section"> <div class="df-shead"> <span class="df-snum">03</span> <span class="df-stitle">Uses</span> </div> <div class="df-sbody"> <h3 class="df-sub">Direct Use</h3> <div class="df-panel" style="margin-bottom:16px;"> <div class="df-row"><span class="df-v">Image-based content moderation classification for Indonesian social media. Given a screenshot, the model produces a structured analysis with a classification label (<code>netral</code>, <code>disinformasi</code>, <code>fitnah</code>, or <code>ujaran kebencian</code>) and a detailed reasoning in Indonesian.</span></div> </div> <h3 class="df-sub">Out-of-Scope Use</h3> <div class="df-panel df-panel--warn"> <div class="df-row"><span class="df-v">This model is not intended for general-purpose vision-language tasks. It is specialized for the DFK disinformation detection pipeline and should not be used for content moderation in other languages or domains without further fine-tuning.</span></div> </div> </div> </div>

<div class="df-section"> <div class="df-shead"> <span class="df-snum">04</span> <span class="df-stitle">Evaluation</span> </div> <div class="df-sbody"> <p>Evaluated on the held-out validation split using greedy decoding (<code>temperature=0.0</code>) and BERTScore (<code>bert-base-multilingual-cased</code>).</p> <div class="df-metrics"> <div class="df-metric df-metric--hi"> <div class="df-mval">94.3</div> <div class="df-mlbl">Accuracy</div> </div> <div class="df-metric"> <div class="df-mval">91.6</div> <div class="df-mlbl">F1 Macro</div> </div> <div class="df-metric"> <div class="df-mval">94.3</div> <div class="df-mlbl">F1 Weighted</div> </div> <div class="df-metric"> <div class="df-mval">80.2</div> <div class="df-mlbl">BERTScore F1</div> </div> </div> <details> <summary>Per-Class Breakdown</summary> <div class="df-drop-body"> <div class="df-panel" style="margin-bottom:0;"> <div class="df-row"><span class="df-k">Netral</span><span class="df-v">P 0.937 · R 0.973 · F1 0.954 · n=970</span></div> <div class="df-row"><span class="df-k">Ujrn Kbnci</span><span class="df-v">P 0.979 · R 0.960 · F1 0.969 · n=867</span></div> <div class="df-row"><span class="df-k">Disinfo</span><span class="df-v">P 0.946 · R 0.895 · F1 0.920 · n=392</span></div> <div class="df-row"><span class="df-k">Fitnah</span><span class="df-v">P 0.822 · R 0.822 · F1 0.822 · n=213</span></div> </div> </div> </details> <details> <summary>Generation Quality Metrics</summary> <div class="df-drop-body"> <div class="df-panel"> <div class="df-phead">BERTScore · bert-base-multilingual-cased</div> <div class="df-row"><span class="df-k">Precision</span><span class="df-v">0.804</span></div> <div class="df-row"><span class="df-k">Recall</span><span class="df-v">0.801</span></div> <div class="df-row"><span class="df-k">F1</span><span class="df-v">0.802</span></div> </div> <div class="df-panel" style="margin-top:12px;"> <div class="df-phead">ROUGE-L · n-gram overlap</div> <div class="df-row"><span class="df-k">Precision</span><span class="df-v">0.400</span></div> <div class="df-row"><span class="df-k">Recall</span><span class="df-v">0.387</span></div> <div class="df-row"><span class="df-k">F1</span><span class="df-v">0.387</span></div> </div> </div> </details> </div> </div>

<div class="df-section"> <div class="df-shead"> <span class="df-snum">05</span> <span class="df-stitle">Training Details</span> </div> <div class="df-sbody"> <h3 class="df-sub">Training Data</h3> <div class="df-panel" style="margin-bottom:16px;"> <div class="df-row"><span class="df-k">Dataset</span><span class="df-v"><code>dfkvlmdatasetv3</code> (augmented on <code>fitnah</code> class)</span></div> <div class="df-row"><span class="df-k">Splits</span><span class="df-v">Fixed (trainaug.csv / valaug.csv)</span></div> <div class="df-row"><span class="df-k">Train</span><span class="df-v">14,293 samples</span></div> <div class="df-row"><span class="df-k">Val</span><span class="df-v">2,831 samples</span></div> </div> <h3 class="df-sub">Label Classes</h3> <div class="df-panel" style="margin-bottom:16px;"> <div class="df-row"><span class="df-k">Netral</span><span class="df-v">Factual content or non-DFK material — no violation detected</span></div> <div class="df-row"><span class="df-k">Disinfo</span><span class="df-v">Claims that contradict established facts, not directed at a specific person</span></div> <div class="df-row"><span class="df-k">Fitnah</span><span class="df-v">False claims directed at a specific individual (defamation)</span></div> <div class="df-row"><span class="df-k">Ujrn Kbnci</span><span class="df-v">Hate speech targeting ethnicity, religion, race, or intergroup identity (SARA)</span></div> </div> <details> <summary>Dataset Distribution</summary> <div class="df-drop-body"> <div class="df-panel"> <div class="df-phead">Train (augmented) · 14,293 total</div> <div class="df-row"><span class="df-k">Netral</span><span class="df-v">3,883 (27.2%)</span></div> <div class="df-row"><span class="df-k">Fitnah</span><span class="df-v">3,846 (26.9%)</span></div> <div class="df-row"><span class="df-k">Ujrn Kbnci</span><span class="df-v">3,484 (24.4%)</span></div> <div class="df-row"><span class="df-k">Disinfo</span><span class="df-v">3,080 (21.6%)</span></div> </div> <div class="df-panel" style="margin-top:12px;"> <div class="df-phead">Val (augmented) · 2,831 total</div> <div class="df-row"><span class="df-k">Netral</span><span class="df-v">970 (34.3%)</span></div> <div class="df-row"><span class="df-k">Ujrn Kbnci</span><span class="df-v">867 (30.6%)</span></div> <div class="df-row"><span class="df-k">Disinfo</span><span class="df-v">765 (27.0%)</span></div> <div class="df-row"><span class="df-k">Fitnah</span><span class="df-v">229 (8.1%)</span></div> </div> </div> </details> <h3 class="df-sub">Configuration</h3> <div class="df-grid" style="margin-top:20px;"> <div class="df-panel"> <div class="df-phead">LoRA Configuration</div> <div class="df-row"><span class="df-k">r</span><span class="df-v">16</span></div> <div class="df-row"><span class="df-k">Alpha</span><span class="df-v">16</span></div> <div class="df-row"><span class="df-k">Dropout</span><span class="df-v">0.1</span></div> <div class="df-row"><span class="df-k">Targets</span><span class="df-v">all-linear</span></div> <div class="df-row"><span class="df-k">Vision</span><span class="df-v">✓ finetuned</span></div> <div class="df-row"><span class="df-k">Language</span><span class="df-v">✓ finetuned</span></div> <div class="df-row"><span class="df-k">Attention</span><span class="df-v">✓ finetuned</span></div> <div class="df-row"><span class="df-k">MLP</span><span class="df-v">✓ finetuned</span></div> </div> <div class="df-panel"> <div class="df-phead">Hyperparameters</div> <div class="df-row"><span class="df-k">Epochs</span><span class="df-v">3</span></div> <div class="df-row"><span class="df-k">Batch</span><span class="df-v">16 (4 × 4 accum)</span></div> <div class="df-row"><span class="df-k">LR</span><span class="df-v">5e-4</span></div> <div class="df-row"><span class="df-k">Optimizer</span><span class="df-v">AdamW 8-bit</span></div> <div class="df-row"><span class="df-k">Max len</span><span class="df-v">4096</span></div> <div class="df-row"><span class="df-k">Grad norm</span><span class="df-v">1</span></div> <div class="df-row"><span class="df-k">Warmup</span><span class="df-v">0.03</span></div> <div class="df-row"><span class="df-k">Grad ckpt</span><span class="df-v">unsloth</span></div> <div class="df-row"><span class="df-k">Seed</span><span class="df-v">3407</span></div> </div> </div> <h3 class="df-sub" style="margin-top:24px;">Trainer</h3> <div class="df-panel" style="margin-bottom:16px;"> <div class="df-row"><span class="df-k">Type</span><span class="df-v"><code>unslothvlmsft</code> (Unsloth VLM SFT trainer)</span></div> <div class="df-row"><span class="df-k">Train on</span><span class="df-v">Responses only</span></div> <div class="df-row"><span class="df-k">Instr part</span><span class="df-v"><code>[INST]</code></span></div> <div class="df-row"><span class="df-k">Resp part</span><span class="df-v"><code>[/INST]</code></span></div> <div class="df-row"><span class="df-k">Best model</span><span class="df-v">Selected by <code>evalloss</code> (lower is better)</span></div> </div> <details> <summary>Prompt Template</summary> <div class="df-drop-body"> <p>Each sample is formatted as a multi-turn conversation using the <code>ministral3</code> chat template. The dataset builds structured content blocks which the Jinja template renders as:</p> <pre><code><s>[SYSTEMPROMPT]...default Ministral system prompt...[/SYSTEM_PROMPT][INST]Anda adalah seorang analis konten media sosial ahli. Diberikan tangkapan layar dari sebuah konten, tentukan label kategori pelanggaran dan berikan analisis detail mengenai pelanggaran yang ditemukan.Ringkasan: {ringkasan} Klaim: {klaim} Fakta: {fakta}[IMG][/INST]Label: {label}

Analisis: {analisis}</s></code></pre>

<h3 class="df-sub" style="margin-top:20px;">Input Fields</h3> <div class="df-panel" style="margin-bottom:16px;"> <div class="df-row"><span class="df-k">Ringkasan</span><span class="df-v">Content summary. In the RAG pipeline this is the concatenation of the image caption (from a captioning model) and any user-provided text (e.g. post caption, tweet text). Effectively holds all available textual context about the content.</span></div> <div class="df-row"><span class="df-k">Klaim</span><span class="df-v">The core claim extracted from the content, used as a web search query for fact-checking. Generated by an LLM from the ringkasan. Can also be a direct caption or user-provided text in simpler setups.</span></div> <div class="df-row"><span class="df-k">Fakta</span><span class="df-v">Verification context retrieved via web search. Contains numbered search results with titles, descriptions, and source URLs. If no relevant sources are found, defaults to <code>"Tidak ditemukan sumber yang valid."</code></span></div> <div class="df-row"><span class="df-k">[IMG]</span><span class="df-v">Screenshot of the social media post being analyzed.</span></div> </div> <h3 class="df-sub">Output Fields</h3> <div class="df-panel"> <div class="df-row"><span class="df-k">Label</span><span class="df-v">One of <code>netral</code>, <code>disinformasi</code>, <code>fitnah</code>, or <code>ujaran kebencian</code>.</span></div> <div class="df-row"><span class="df-k">Analisis</span><span class="df-v">Free-form Indonesian-language explanation of why the content was assigned its label, referencing the image, context, and any retrieved facts.</span></div> </div> </div> </details> <details> <summary>Full Training Config</summary> <div class="df-drop-body"> <pre><code>experiment_name: final-ministral-8b-cpt-ws3 seed: 3407

reporting: wandb: true wandb_project: "DFK3"

model: name: unslothvlm pretrained: aitf-komdigi/KomdigiITS-8B-DFK-CPT kwargs: loadin4bit: false chattemplate: "sita/templates/ministral_3.jinja"

adapter: name: unslothvlmlora kwargs: finetunevisionlayers: true finetunelanguagelayers: true finetuneattentionmodules: true finetunemlpmodules: true r: 16 loraalpha: 16 loradropout: 0.1 bias: "none" targetmodules: "all-linear" usegradientcheckpointing: "unsloth" randomstate: 3407

dataset: name: dfkvlmdatasetv3 kwargs: datadir: /content/dataset/images/images

training: numepochs: 3 batchsize: 4 learningrate: 5e-4 gradientaccumulationsteps: 4 maxgradnorm: 1 warmupratio: 0.03 weightdecay: 0 loggingsteps: 1 evalsteps: 250 extra: seed: 3407 maxlength: 4096 loadbestmodelatend: true metricforbestmodel: evalloss greaterisbetter: false

trainer: name: unslothvlmsft kwargs: trainonresponsesonly: true instructionpart: "[INST]" responsepart: "[/INST]" optim: adamw8bit

evaluation: name: vlmgen kwargs: maxnewtokens: 512 temperature: 0.0 bertmodel: bert-base-multilingual-cased batchsize: 16 numworkers: 11</code></pre>

</div> </details> </div> </div>

<div class="df-section"> <div class="df-shead"> <span class="df-snum">06</span> <span class="df-stitle">Model Sources</span> </div> <div class="df-sbody"> <div class="df-panel"> <div class="df-row"><span class="df-k">Framework</span><span class="df-v"><a href="https://github.com/aitf-its-tim3-dfk/SITA">SITA</a></span></div> <div class="df-row"><span class="df-k">W&B Run</span><span class="df-v"><a href="https://wandb.ai/aitfits2026-kementerian-komdigi/DFK3/runs/zbplofhf">DFK3 / final-ministral-8b-cpt-ws3</a></span></div> </div> </div> </div>

<div class="df-section"> <div class="df-shead"> <span class="df-snum">07</span> <span class="df-stitle">Framework Versions</span> </div> <div class="df-sbody"> <div class="df-panel"> <div class="df-row"><span class="df-k">TRL</span><span class="df-v">0.24.0</span></div> <div class="df-row"><span class="df-k">Transformers</span><span class="df-v">5.5.0</span></div> <div class="df-row"><span class="df-k">PyTorch</span><span class="df-v">2.11.0+cu128</span></div> <div class="df-row"><span class="df-k">Datasets</span><span class="df-v">4.3.0</span></div> <div class="df-row"><span class="df-k">PEFT</span><span class="df-v">0.19.0</span></div> <div class="df-row"><span class="df-k">Tokenizers</span><span class="df-v">0.22.2</span></div> </div> </div> </div>

</div> </body> </html>