{"repo_id": "d0rj/q-51M-base", "arch_family": "causal-gpt", "arch_note": "reference causal baseline", "step": 15000, "source_tokens": 3932160000, "arch": {"n_layers": 10, "d_model": 512, "n_heads": 8, "kv_heads": 2, "ffn": 1792, "vocab": 32768, "ctx": 2048, "tie_emb": true, "model_type": "q50m", "architectures": ["Q50MForCausalLM"], "extra": {"head_dim": 64}}, "scores": {"hellaswag": {"metric": "acc_norm", "value": 0.291775, "ci95": [0.282965, 0.300744], "n": 10042}, "arc_easy": {"metric": "acc_norm", "value": 0.433081, "ci95": [0.413281, 0.453097], "n": 2376}, "arc_challenge": {"metric": "acc_norm", "value": 0.242321, "ci95": [0.218657, 0.267668], "n": 1172}, "piqa": {"metric": "acc_norm", "value": 0.599021, "ci95": [0.576431, 0.621197], "n": 1838}, "winogrande": {"metric": "acc", "value": 0.500395, "ci95": [0.472904, 0.527883], "n": 1267}, "openbookqa": {"metric": "acc_norm", "value": 0.282, "ci95": [0.244336, 0.322988], "n": 500}, "boolq": {"metric": "acc", "value": 0.598777, "ci95": [0.581871, 0.615451], "n": 3270}, "lambada_openai": {"metric": "acc", "value": 0.208616, "ci95": [0.197741, 0.219925], "n": 5153}}, "ar_macro": 0.394498, "metric_type": "AR-loglikelihood", "duplicate_of": null}
{"repo_id": "d0rj/q-prefixlm-51M-base", "arch_family": "prefix-lm", "arch_note": "bidirectional prefix, suffix-only loss", "step": 15000, "source_tokens": 3932160000, "arch": {"n_layers": 10, "d_model": 512, "n_heads": 8, "kv_heads": 2, "ffn": 1792, "vocab": 32768, "ctx": 2048, "tie_emb": true, "model_type": "prefix_lm", "architectures": ["PrefixLMForCausalLM"], "extra": {"head_dim": 64, "training_objective": "suffix-only-v2"}}, "scores": {"hellaswag": {"metric": "acc_norm", "value": 0.283908, "ci95": [0.275173, 0.292808], "n": 10042}, "arc_easy": {"metric": "acc_norm", "value": 0.362374, "ci95": [0.343282, 0.38191], "n": 2376}, "arc_challenge": {"metric": "acc_norm", "value": 0.227816, "ci95": [0.204715, 0.252695], "n": 1172}, "piqa": {"metric": "acc_norm", "value": 0.531012, "ci95": [0.508157, 0.553738], "n": 1838}, "winogrande": {"metric": "acc", "value": 0.497238, "ci95": [0.469756, 0.524735], "n": 1267}, "openbookqa": {"metric": "acc_norm", "value": 0.256, "ci95": [0.219708, 0.296013], "n": 500}, "boolq": {"metric": "acc", "value": 0.548624, "ci95": [0.531521, 0.565613], "n": 3270}, "lambada_openai": {"metric": "acc", "value": 0.233456, "ci95": [0.222107, 0.245202], "n": 5153}}, "ar_macro": 0.367553, "metric_type": "AR-loglikelihood", "duplicate_of": null}
{"repo_id": "d0rj/prefixlm-51M-base", "arch_family": "prefix-lm", "arch_note": "prefix-LM (this repo id is the prefixlm variant)", "step": 15000, "source_tokens": 3932160000, "arch": {"n_layers": 10, "d_model": 512, "n_heads": 8, "kv_heads": 2, "ffn": 1792, "vocab": 32768, "ctx": 2048, "tie_emb": true, "model_type": "prefix_lm", "architectures": ["PrefixLMForCausalLM"], "extra": {"head_dim": 64, "training_objective": "suffix-only-v2"}}, "scores": {"hellaswag": {"metric": "acc_norm", "value": 0.283908, "ci95": [0.275173, 0.292808], "n": 10042}, "arc_easy": {"metric": "acc_norm", "value": 0.362374, "ci95": [0.343282, 0.38191], "n": 2376}, "arc_challenge": {"metric": "acc_norm", "value": 0.227816, "ci95": [0.204715, 0.252695], "n": 1172}, "piqa": {"metric": "acc_norm", "value": 0.531012, "ci95": [0.508157, 0.553738], "n": 1838}, "winogrande": {"metric": "acc", "value": 0.497238, "ci95": [0.469756, 0.524735], "n": 1267}, "openbookqa": {"metric": "acc_norm", "value": 0.256, "ci95": [0.219708, 0.296013], "n": 500}, "boolq": {"metric": "acc", "value": 0.548624, "ci95": [0.531521, 0.565613], "n": 3270}, "lambada_openai": {"metric": "acc", "value": 0.233456, "ci95": [0.222107, 0.245202], "n": 5153}}, "ar_macro": 0.367553, "metric_type": "AR-loglikelihood", "duplicate_of": "d0rj/q-prefixlm-51M-base"}
{"repo_id": "d0rj/looped-51M-base", "arch_family": "looped", "arch_note": "shared blocks x loops (Universal-Transformer style)", "step": 15000, "source_tokens": 3932160000, "arch": {"n_layers": 60, "d_model": 512, "n_heads": 8, "kv_heads": 2, "ffn": 1792, "vocab": 32768, "ctx": 2048, "tie_emb": true, "model_type": "looped_lm", "architectures": ["LoopedLMForCausalLM"], "extra": {"head_dim": 64}}, "scores": {"hellaswag": {"metric": "acc_norm", "value": 0.296156, "ci95": [0.287306, 0.305162], "n": 10042}, "arc_easy": {"metric": "acc_norm", "value": 0.442761, "ci95": [0.422897, 0.46281], "n": 2376}, "arc_challenge": {"metric": "acc_norm", "value": 0.22099, "ci95": [0.198168, 0.245634], "n": 1172}, "piqa": {"metric": "acc_norm", "value": 0.602829, "ci95": [0.580267, 0.624962], "n": 1838}, "winogrande": {"metric": "acc", "value": 0.501184, "ci95": [0.473691, 0.52867], "n": 1267}, "openbookqa": {"metric": "acc_norm", "value": 0.29, "ci95": [0.251947, 0.331255], "n": 500}, "boolq": {"metric": "acc", "value": 0.615902, "ci95": [0.599105, 0.632428], "n": 3270}, "lambada_openai": {"metric": "acc", "value": 0.209004, "ci95": [0.198122, 0.220321], "n": 5153}}, "ar_macro": 0.397353, "metric_type": "AR-loglikelihood", "duplicate_of": null}
{"repo_id": "d0rj/diffusion-51M-base", "arch_family": "diffusion", "arch_note": "masked/diffusion LM - DIFFERENT METRIC, excluded from AR comparison", "step": 15000, "source_tokens": 3932160000, "arch": {"n_layers": 10, "d_model": 512, "n_heads": 8, "kv_heads": null, "ffn": 1536, "vocab": 32768, "ctx": 2048, "tie_emb": true, "model_type": "diffusion_lm", "architectures": ["DiffusionLMForMaskedLM"], "extra": {"head_dim": 64, "training_objective": "absorbing-linear-1overT-v2"}}, "scores": {"hellaswag": {"metric": "acc_norm", "value": 0.279526, "ci95": [0.270834, 0.288386], "n": 10042}, "arc_easy": {"metric": "acc_norm", "value": 0.336279, "ci95": [0.317561, 0.355526], "n": 2376}, "arc_challenge": {"metric": "acc_norm", "value": 0.222696, "ci95": [0.199804, 0.2474], "n": 1172}, "piqa": {"metric": "acc_norm", "value": 0.538629, "ci95": [0.515782, 0.561315], "n": 1838}, "winogrande": {"metric": "acc", "value": 0.501184, "ci95": [0.473691, 0.52867], "n": 1267}, "openbookqa": {"metric": "acc_norm", "value": 0.26, "ci95": [0.223486, 0.300174], "n": 500}, "boolq": {"metric": "acc", "value": 0.529664, "ci95": [0.512532, 0.546726], "n": 3270}, "lambada_openai": {"metric": "acc", "value": 0.422084, "ci95": [0.408662, 0.435622], "n": 5153}}, "ar_macro": null, "metric_type": "diffusion-continuation-PLL", "duplicate_of": null}