messages list | tier int64 | src string |
|---|---|---|
[
{
"role": "user",
"content": "[Request interrupted by user]\n<local-command-caveat>Caveat: The messages below were generated by the user while running local commands. DO NOT respond to these messages or otherwise consider them in your response unless the user explicitly asks you to.</local-command-caveat>\n... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "[Request interrupted by user]\n<local-command-caveat>Caveat: The messages below were generated by the user while running local commands. DO NOT respond to these messages or otherwise consider them in your response unless the user explicitly asks you to.</local-command-caveat>\n... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "[Request interrupted by user]\n<local-command-caveat>Caveat: The messages below were generated by the user while running local commands. DO NOT respond to these messages or otherwise consider them in your response unless the user explicitly asks you to.</local-command-caveat>\n... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "[Request interrupted by user]\n<local-command-caveat>Caveat: The messages below were generated by the user while running local commands. DO NOT respond to these messages or otherwise consider them in your response unless the user explicitly asks you to.</local-command-caveat>\n... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\ntelemetry.progress import emit_log_line\n22\tfrom src.mythosmini.telemetry.runlog import RunLog\n23\tfrom src.mythosmini.training.optimizer import build_nadamw\n24\tfrom src.mythosmini.training.scheduler import …[truncated]\nAssistant action: Read {\"file... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "[Request interrupted by user]\n<local-command-caveat>Caveat: The messages below were generated by the user while running local commands. DO NOT respond to these messages or otherwise consider them in your response unless the user explicitly asks you to.</local-command-caveat>\n... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n(1, int(total_steps * config.warmup_fraction))\n8\t decay_steps = max(1, int(total_steps * config.decay_fraction))\n9\t decay_start = total_steps - decay_steps\n10\t floor = config.peak_lr * config.min_lr_fraction\n11\t if step < warmup_steps:... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n\n10\tfrom src.mythosmini.checkpointing.saver import BackgroundSaver\n11\tfrom src.mythosmini.config import (\n12\t CHECKPOINT_INTERVAL_STEPS,\n13\t LOG_INTERVAL_STEPS,\n14\t ModelConfig,\n15\t TrainConfig,\n16\t)\n17\tfrom src.mythosmini.data... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nles\n21\t levels = torch.tensor(FP4_LEVELS, dtype=normalized.dtype, device=normalized.device)\n22\t midpoints = (levels[:-1] + levels[1:]) / 2\n23\t nearest = levels[torch.bucketize(normalized.abs(), midpoints)]\n24\t quantized = nearest * nor... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n.25\n27\t grad_clip: float = 1.0\n28\t seed: int = 1337\n29\t floor_tighten_per_round: float = 0.05\n30\t f …[truncated]\nAssistant action: Read {\"file_path\": \"/home/user/MythosMini/src/mythosmini/training/optimizer.py\"}\nResult: 1\tfrom t... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n7\tfrom torch.nn import functional\n8\t\n9\tfrom src.mythosmini.config import ModelConfig\n10\tfrom src.mythosmini.training.fp4 import Fp4Linear\n11\t\n12\tLOCAL_WINDOW_BONUS = 1.0e4\n13\tFLEX_MIN_HEAD_DIM = 16\n14\t\n15\t_compiled_flex_attention: Callabl... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nude + 1 shared block looped 2-6x with per-loop LoRA + 4 coda, dim 96, RMSNorm/RoPE/SwiGLU, tied embeddings, DSA-style sparse attention (shared model-level lightning indexer, top-128 keys, 64 local window, FlexAttention short path, chunked long path, …[tru... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n-- src/mythosmini/model/transformer.py | head -250\", \"description\": \"Show diff of shared indexer commit\"}\nResult: commit 28cf7bee2e59ae47a25b114ea4f0383c9e840df0\nAuthor: CompactAI <ai@demo_project.com>\nDate: Thu Jun 11 16:21:17 2026 -0400\n\n ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nwikitext-2-raw-v1\"\n28\t\n29\tBLIMP_PAIRS_PER_SUBTASK = 100\n30\tARC_QUESTION_COUNT = 500\n31\tARC_FEWSHOT_COUNT = 5\n32\tWIKITEXT_MAX_BYTES = 524288\n33\t\n34\tCHECKPOINT_INTERVAL_STEPS = 2000\n35\tLOG_INTERVAL_STEPS = 2000\n36\t\n37\tOOM_BAT …[truncate... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n paperclip incident\ndb3c7a4 record salvaged throughput-restructure run scores\n1061bde persist shrunk stm micro batch across epochs and rounds\ndf79253 fix gemma\n111d3e6 Othr\n6631bf2 fixes\n4b21513 Merge branch 'main' of github.com:CompactAIOfficial/My... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nch.py downloads with cache fallback), hardware/ (detect.py uses torch.cuda.mem_get_info so co-tenant processes like the running ollama are respected, never killed; autotune.py …[truncated]\nAssistant action: Read {\"file_path\": \"/home/user/MythosMini/s... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nhosMini merge\nb475cd0 catch oom in stm backward pass too\n58fb61f auto-shrink micro batch on oom in stm and sft loops\na6559c4 journal throughput optimization work\n28cf7be shared sparse indexer with flex attention path for 1.8x throughput\n3668a97 journ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n seq_len: int = 512\n/home/user/MythosMini/src/mythosmini/config.py:80: peak_lr: float = 3e-3\n/home/user/MythosMini/src/mythosmini/config.py:82: warmup_fraction: float = 0.01\n/home/user/MythosMini/src/mythosmini/config.py:84: weight_decay: floa... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n.png\nhero_restored.png\nissues.json\njournal.md\nMakefile\nmessage.txt\nnode_modules\noptimizations.md\npackage.json\npackage-lock.json\nplan.md\npyaot.py\n__pycache__\npyproject.toml\nralph\nralph.py\nREADME.md\nrepro_data.py\nrequirements.txt\nresearch... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n\nloop.py\noptimizer.py\n__pycache__\nscheduler.py\nsft.py\nstm.py\nAssistant action: Bash {\"command\": \"sed -n 1,120p /home/user/MythosMini/src/mythosmini/config.py\", \"description\": \"Read config file\"}\nResult: from dataclasses import dataclass, f... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\ne/MythosMini/src/mythosmini/model:\n__init__.py\n__pycache__\ntransformer.py\n\n/home/user/MythosMini/src/mythosmini/training:\nfp4.py\n__init__.py\nloop.py\noptimizer.py\n__pycache__\nscheduler.py\nsft.py\nstm.py\nAssistant action: Bash {\"command\": \"s... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nnoGPT Speedrun Optimizations | Evan Conway\",\"url\":\"https://evanjayconway.com/posts/2026/nanogpt-improvements/\"},{\"title\":\"Muon Optimizer Implementation\",\"url\":\"https://www.emergentmind.com/topics/muon-optimizer-implementation\"},{\"title\":\"G... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nBatch Size in LLM Pre-training\",\"url\":\"https://arxiv.org/pdf/2505.13738\"},{\"title\":\"Theory of Optimal Learning Rate Schedules and Scaling Laws for a Random\",\"url\":\"https://arxiv.org/pdf/2602.04774\"},{\"title\":\"Functional Scaling Laws in Ker... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nne Learning\",\"url\":\"https://www.emergentmind.com/topics/model-scaling-laws\"},{\"title\":\"Efficient Hyperparameter Tuning via Trajectory Invariance Principle\",\"url\":\"https://arxiv.org/pdf/2509.25049\"},{\"title\":\"Power Lines: Scaling Laws for W... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nptimizer small language model training speedrun modded-nanogpt results 2025\"\n\nLinks: [{\"title\":\"Muon: An optimizer for hidden layers in neural networks | Keller Jordan blog\",\"url\":\"https://kellerjordan.github.io/posts/muon/\"},{\"title\":\"The A... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nn neural networks | Keller Jordan blog\",\"url\":\"https://kellerjordan.github.io/posts/muon/\"},{\"title\":\"The Automated LLM Speedrunning Benchmark: Reproducing NanoGPT Improvements\",\"url\":\"https://openreview.net/pdf?id=w98hMEjzu8\"},{\"title\":\"A... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nuery: \"optimal learning rate scaling law model size data size power law 2025 predictable scale step law\"\n\nLinks: [{\"title\":\"Scaling Laws in Machine Learning\",\"url\":\"https://www.emergentmind.com/topics/model-scaling-laws\"},{\"title\":\"Efficien... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\ntillation from an ensemble of teachers trained on a small dataset with no performance penalty\",\"url\":\"https://arxiv.org/abs/2308.02019\"},{\"title\":\"[2308.02019v2] Baby Llama: knowledge distillation from an ensemble of teachers trained on a small da... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nBenchmark: Reproducing NanoGPT Improvements\",\"url\":\"https://openreview.net/pdf?id=w98hMEjzu8\"},{\"title\":\"A Field Guide to NanoGPT Speedrun Optimizations | Evan Conway\",\"url\":\"https://evanjayconway.com/posts/2026/nanogpt-improvements/\"},{\"tit... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nn language model\"}\nResult: Web search results for query: \"softmax bottleneck small language model saturation mixture of softmaxes hidden dimension rank\"\n\nLinks: [{\"title\":\"[2404.07647] Why do small language models underperform? Studying Language ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nck\",\"url\":\"https://arxiv.org/abs/2404.07647\"},{\"title\":\"Why do small language models underperform? Studying LM Saturation via the Softmax Bottleneck\",\"url\":\"https://arxiv.org/html/2404.07647v1\"},{\"title\":\"Why do small language models under... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nvariance Principle\",\"url\":\"https://arxiv.org/pdf/2509.25049\"},{\"title\":\"Power Lines: Scaling Laws for Weight Decay and Batch Size in LLM Pre-training\",\"url\":\"https://arxiv.org/pdf/2505.13738\"},{\"title\":\"Theory of Optimal Learning Rate Sche... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nion from an ensemble of teachers trained on a small dataset with no performance penalty | Papers with Code\",\"url\":\"https://paperswithcode.co/paper/2308.02019\"},{\"title\":\"Paper page - …[truncated]\nResult: Web search results for query: \"extreme ov... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n\"Baby Llama knowledge distillation ensemble teachers small dataset BabyLM\"\n\nLinks: [{\"title\":\"Baby Llama: knowledge distillation from an ensemble of teachers trained on a small dataset with no performance penalty | Takara TLDR\",\"url\":\"https://t... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nh size is primarily influenced by D and remains largely invariant to N,\" but the actual mathematical formulas with numerical coefficients are not included in this webpage excerpt.\n\nTo obtain the precise formulas, you would need to access the full PDF o... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nquality improvements\n- Operating with high tokens-per-parameter ratios\n\nThe …[truncated]\nResult: Web search results for query: \"multi-token prediction auxiliary loss small language model byte level dense supervision improvement\"\n\nLinks: [{\"title\... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n/abs/2404.19737\", \"prompt\": \"What does this paper report about multi-token prediction for byte-level models specifically? How many bytes ahead were predicted and what improvement was seen on benchmarks for byte-level training?\"}\nAssistant action: We... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\ng.py:82: warmup_fraction: float = 0.01\n/home/user/MythosMini/src/mythosmini/config.py:84: weight_decay: float = 0.01\n---\n/home/user/MythosMini/src/mythosmini/model:\n__init__.py\n__pycache__\ntransformer.py\n\n/home/user/MythosMini/src/mythosmini... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n5\timport numpy as np\n6\timport torch\n7\tfrom torch import Tensor\n8\tfrom torch.nn import functional\n9\t\n10\tfrom src.mythosmini.config import ModelConfig, TrainConfig\n11\tfrom src.mythosmini.data.cache import CACHE_DTYPE\n12\tfrom src.mythosmini.ha... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nZED only logarithmically — doubling params costs ~5% efficiency but any overall gain counts fully). Current pipeline: procedural Dyck warmup -> 2B-token training (NAdamW, WSD schedule) -> 3-round recursive STM -> benchmark all checkpoints -> SLERP tournam... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n.transformer import MythosMini\n7\t\n8\t\n9\t@dataclass(frozen=True)\n10\tclass ScoredPair:\n11\t context: list[int]\n12\t continuation: list[int]\n13\t\n14\t\n15\tdef truncate_pair(pair: ScoredPair, max_len: int) -> tuple[list[int], int]:\n16\t ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\ndigm\",\"url\":\"https://www.emergentmind.com/topics/warmup-stable-decay-wsd\"},{\"title\":\"learning rate scheduling: warmup-stable-decay vs cosine – Ozan Ciga\",\"url\":\"https://ozanciga.wordpress.com/2025/10/06/learning-rate-scheduling-warmup-stable-d... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nmport Tensor\n6\t\n7\tfrom src.mythosmini.benchmarks.runner import BenchmarkScores\n8\tfrom src.mythosmini.config import SLERP_MAX_GENERATIONS, SLERP_T_SWEEP, SLERP_TOP_K\n9\tfrom src.mythosmini.merging.slerp import slerp_state_dicts\n10\t\n11\tSTALL_GENE... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nh\n6\tfrom torch import Tensor\n7\tfrom torch.nn import functional\n8\tfrom tqdm import tqdm\n9\t\n10\tfrom src.mythosmini.checkpointing.saver import BackgroundSaver\n11\tfrom src.mythosmini.config import (\n12\t CHECKPOINT_INTERVAL_STEPS,\n13\t LOG... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n: ModelConfig) -> type[nn.Linear]:\n28\t return Fp4Linear if config.fp4_weights else nn.Linear\n29\t\n30 …[truncated]\nAssistant action: Read {\"file_path\": \"/home/user/MythosMini/src/mythosmini/benchmarks/wikitext.py\"}\nResult: 1\timport math\n2\t\... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nuracy, load_arc_questions\n6\tfrom src.mythosmini.benchmarks.blimp import blimp_accuracy, load_blimp_pairs\n7\tfrom src.mythosmini.benchmarks.wikitext import byte_perplexity, load_wikitext_bytes\n8\tfrom src.mythosmini.hardware.autotune import autotune_ba... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\ntle\":\"WSD Learning Rate Scheduling\",\"url\":\"https://www.emergentmind.com/topics/warmup-stable-decay-wsd-learning-rate-scheduling\"},{\"title\":\"WSD Schedules for Efficient Deep Learning\",\"url\":\"https://www.emergentmind.com/topics/warmup-stable-d... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nni/src/mythosmini/training/fp4.py\n/home/user/MythosMini/src/mythosmini/training/stm.py\n/home/user/MythosMini/src/mythosmini/training/scheduler.py\n/home/user/MythosMini/src/mythosmini/training/__init__.py\n/home/user/MythosMini/src/mythosmini/training/s... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nn labels.index(answer_key)\n24\t\n25\t\n26\tdef record_to_question(record: dict[str, Any]) -> ArcQuestion | None:\n27\t index = answer_index_of(record)\n28\t if index is None:\n29\t return None\n30\t …[truncated]\nAssistant action: Read {... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\ny\n/home/user/MythosMini/src/mythosmini/training/stm.py\n/home/user/MythosMini/stm.py\n/home/user/MythosMini/tests/e2e/test_train_entry.py\n/home/user/MythosMini/tests/test_merging.py\n/home/user/MythosMini/tests/test_stm.py\n/home/user/MythosMini/train.p... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\naset(BLIMP_DATASET, name=config_name, split=\"train\")\n16\t pairs: list[tuple[str, str]] = []\n17\t for record in dataset:\n18\t pairs.append((str(record[\"sentence_good\"]), str(record[\"sentence_bad\"])))\n19\t if le... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\ntchPlan:\n18\t micro_batch: int\n19\t grad_accum: int\n20\t\n21\t @property\n22\t def effective_batch(self) -> int:\n23\t return self.micro_batch * self.grad_accum\n24\t\n25\t\n26\tdef shrink_after_oom(plan: BatchPlan) -> BatchPlan:\n27... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\ntchPlan(micro_batch=new_micro, grad_accum=new_accum)\n30\t\n31\t\n32\tdef time_probe_steps(step_fn: Callable[[int], None], batch_si …[truncated]\nAssistant action: Read {\"file_path\": \"/home/user/MythosMini/src/mythosmini/training/fp4.py\"}\nResult: 1\t... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nurrent pipeline: procedural Dyck warmup -> 2B-token training (NAdamW, WSD schedule) -> 3-round recursive STM -> benchmark all checkpoints -> SLERP tournament -> SFT instruct phase -> instruct SLERP. Architecture: 4 prelude + 1 shared block looped 2-6x wit... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nstruct SLERP. Architecture: 4 prelude + 1 shared block looped 2-6x with per-loop LoRA + 4 coda, dim 96, RMSNorm/RoPE/SwiGLU, tied embeddings, DSA-style sparse attention (shared model-level lightning indexer, top-128 keys, 64 local window, FlexAttention sh... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n\n\nLinks: [{\"title\":\"README.md · roneneldan/TinyStories at main\",\"url\":\"https://huggingface.co/datasets/roneneldan/TinyStories/blob/main/README.md\"},{\"title\":\"[2305.07759] TinyStories: How Small Can Language Models Be and Still Speak Coherent ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n/article/view/34543/36698\"},{\"title\":\"DWA-KD: Dual-Space Weighting and Time-Warped Alignment for Cross-Tokenizer Knowledge Distillation\",\"url\":\"https://arxiv …[truncated]\nAssistant action: WebFetch {\"url\": \"https://arxiv.org/html/2604.07466v1\... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n Sample-Efficient Pretraining on Developmentally Plausible Corpora | Rycolab\",\"url\":\"https://rycolab.io/publication/warstadtal-babylm-23/\"},{\"title\":\"GitHub - babylm/evaluation-pipeline-2024: The evaluation pipeline for the 2024 BabyLM Challenge. ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\ne in (Large) Language Models\",\"url\":\"https://arxiv.org/pdf/2405.17767\"},{\"title\":\"README.md · abhilash88/tinystories-slm-gpt at e83d515e8722034f455656c742f2debb76e9f412\",\"url\":\"https://huggingface.co/abhilash88/tinystories-slm-gpt/blob/e83d515... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n - babylm/evaluation-pipeline-2024: The evaluation pipeline for the 2024 BabyLM Challenge. · GitHub\",\"url\":\"https://github.com/babylm/evaluation-pipeline-2024\"},{\"title\":\"BabyLM Challenge Overview\",\"url\":\"https://www.emergentmind.com/topics/ba... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nhmarks for LLM Inference: RTX, H100, B200\",\"url\":\"https://www.cloudrift.ai/gpu-benchmarks\"},{\"title\":\"The RTX 5090 Is Here: Serve 65,000+ Tokens per Second on RunPod\",\"url\": …[truncated]\nAssistant action: WebSearch {\"query\": \"BabyLM 2025 fi... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n\"[2305.07759] TinyStories: How Small Can Language Models Be and Still Speak Coherent English?\",\"url\":\"https://arxiv.org/abs/2305.07759\"},{\"title\":\"TinyStories: How Small Can Language Models Be and Still Speak Coherent English?\",\"url\":\"https:/... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n across the dataset\n- Training split: \"2.12M rows\"\n- Test split: \"21.4k rows\"\n- Format: Parquet files\n\n**Language:**\nEnglish only\n\n**License:**\nMIT license\n\n**Generation Model:**\n\"gpt-4o-mini-2024-07-18\" (all visible examples use this mo... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nfrom collections.abc import Iterator\nfrom typing import Any\n\nimport torch\nfrom torch import Tensor\nfrom torch.utils.data import DataLoader, IterableDataset, get_worker_info\n\nfrom src.mythosmini.config import FINEWEB_CONFIG, FINEWEB_DATASET, LOADER_... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\nthosMini/src/mythosmini/data/*.py 2>/dev/null | head -120\", \"description\": \"Inspect data pipeline module\"}\nResult: buffer.py\ncache.py\nfineweb.py\n__init__.py\npacking.py\nprocedural.py\n__pycache__\n==> /home/user/MythosMini/src/mythosmini/data/bu... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\n Generation Code for SimpleStories\",\"url\":\"https://github.com/lennart-finke/simple_stories_generate\"},{\"title\":\"GitHub - simple-stories/simple_stories_generate: Dataset Generation Code for SimpleStories · GitHub\",\"url\":\"https://github.com/simp... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "Repo: /home/user/MythosMini — a 1,050,241-param byte-level (vocab 257) looped-shared-block transformer trained on fineweb-edu, benchmarked on WikiText-2 byte PPL, BLiMP, ARC-Easy against the Tiny-ML leaderboard (overall = mean of blimp, arc, normalized wiki score; efficiency = ... | 1 | Swarm-AI-Research/fable5-traces-sft |
[
{
"role": "user",
"content": "…[earlier truncated]…\no-mini-2024-07-18\" (all visible examples use this model)\n\n**Annotation Fields:**\nThe dataset includes rich metadata columns:\n- Story content and narrative attributes (topic, theme, style, feature, persona)\n- Grammar specifications (grammar, initial_... | 1 | Swarm-AI-Research/fable5-traces-sft |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.