v1: 6-model same-harness arch→score panel (lm-eval 0.4.12, 0-shot, seed 1234, decontaminated)
Browse files- rows.jsonl +6 -0
rows.jsonl
ADDED
|
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{"model_id":"harrrshall/BarunLM-35M","params":35072768,"family":"from-scratch","arch":"hybrid_local_full_attn + selective_residual_routing","layers":12,"d_model":448,"n_heads":7,"n_kv_heads":1,"ffn_dim":1228,"ffn_act":"swiglu","vocab":16384,"ctx":2048,"pos_enc":"rope","norm":"rmsnorm","tie_emb":true,"attn_types":["local","local","local","full"],"training_tokens":5699985408,"tok_per_param":162.5,"macro_accuracy":0.4101019372910188,"arc_challenge":0.222990492653414,"arc_easy":0.38862761864044465,"boolq":0.5593303791235844,"hellaswag":0.2899662473151273,"lambada_openai":0.2450731707317073,"openbookqa":0.28,"piqa":0.5926950747094631,"sciq":0.5992115637319316,"winogrande":0.5130228887134964,"harness":"lm-eval==0.4.12","num_fewshot":0,"seed":1234,"decontaminated":true,"scope":"general-domain","source":"harrrshall/BarunLM-35M benchmark_results.json"}
|
| 2 |
+
{"model_id":"LiquidAI/LFM2.5-230M-Base","params":229693184,"family":"semi-big-lab (LiquidAI LFM line)","arch":"lfm2 hybrid conv + full attention","layers":14,"d_model":1024,"n_heads":16,"n_kv_heads":8,"ffn_dim":2560,"ffn_act":"swiglu","vocab":65536,"ctx":128000,"pos_enc":"rope + conv","norm":"rmsnorm","tie_emb":true,"attn_types":["conv","conv","full","conv","full","conv","full","conv","full","conv","full","conv","full","conv"],"training_tokens":null,"tok_per_param":null,"macro_accuracy":0.39197443061078113,"arc_challenge":null,"arc_easy":null,"boolq":null,"hellaswag":null,"lambada_openai":null,"openbookqa":null,"piqa":null,"sciq":null,"winogrande":null,"harness":"lm-eval==0.4.12","num_fewshot":0,"seed":1234,"decontaminated":true,"scope":"general-domain","source":"harrrshall/BarunLM-35M benchmark_results.json (comparison table, macro only)"}
|
| 3 |
+
{"model_id":"EleutherAI/pythia-160m-deduped","params":162322944,"family":"semi-big-lab (EleutherAI pythia line)","arch":"gpt_neox rotary","layers":12,"d_model":768,"n_heads":12,"n_kv_heads":12,"ffn_dim":3072,"ffn_act":"gelu","vocab":50304,"ctx":2048,"pos_enc":"rope_25pct","norm":"layernorm","tie_emb":false,"attn_types":["full"],"training_tokens":null,"tok_per_param":null,"macro_accuracy":0.3734593023210671,"arc_challenge":null,"arc_easy":null,"boolq":null,"hellaswag":null,"lambada_openai":null,"openbookqa":null,"piqa":null,"sciq":null,"winogrande":null,"harness":"lm-eval==0.4.12","num_fewshot":0,"seed":1234,"decontaminated":true,"scope":"general-domain","source":"harrrshall/BarunLM-35M benchmark_results.json (comparison table, macro only)"}
|
| 4 |
+
{"model_id":"StentorLabs/Stentor-30M","params":30419712,"family":"from-scratch","arch":"llama","layers":21,"d_model":256,"n_heads":4,"n_kv_heads":4,"ffn_dim":1024,"ffn_act":"silu","vocab":32768,"ctx":512,"pos_enc":"rope","norm":"rmsnorm","tie_emb":true,"attn_types":["full"],"training_tokens":null,"tok_per_param":null,"macro_accuracy":0.3645857462123749,"arc_challenge":null,"arc_easy":null,"boolq":null,"hellaswag":null,"lambada_openai":null,"openbookqa":null,"piqa":null,"sciq":null,"winogrande":null,"harness":"lm-eval==0.4.12","num_fewshot":0,"seed":1234,"decontaminated":true,"scope":"general-domain","source":"harrrshall/BarunLM-35M benchmark_results.json (comparison table, macro only)"}
|
| 5 |
+
{"model_id":"roneneldan/TinyStories-33M","params":68514048,"family":"narrow-domain diagnostic (TinyStories)","arch":"gpt_neo alternating global/local","layers":4,"d_model":768,"n_heads":16,"n_kv_heads":16,"ffn_dim":3072,"ffn_act":"gelu_new","vocab":50257,"ctx":2048,"pos_enc":"learned","norm":"layernorm","tie_emb":true,"attn_types":["global","local","global","local"],"training_tokens":null,"tok_per_param":null,"macro_accuracy":0.33156687160177734,"arc_challenge":null,"arc_easy":null,"boolq":null,"hellaswag":null,"lambada_openai":null,"openbookqa":null,"piqa":null,"sciq":null,"winogrande":null,"harness":"lm-eval==0.4.12","num_fewshot":0,"seed":1234,"decontaminated":true,"scope":"narrow-domain diagnostic","source":"harrrshall/BarunLM-35M benchmark_results.json (comparison table, macro only)"}
|
| 6 |
+
{"model_id":"EleutherAI/pythia-70m-deduped","params":70426624,"family":"semi-big-lab (EleutherAI pythia line)","arch":"gpt_neox rotary","layers":6,"d_model":512,"n_heads":8,"n_kv_heads":8,"ffn_dim":2048,"ffn_act":"gelu","vocab":50304,"ctx":2048,"pos_enc":"rope_25pct","norm":"layernorm","tie_emb":false,"attn_types":["full"],"training_tokens":null,"tok_per_param":null,"macro_accuracy":0.3171108992671423,"arc_challenge":null,"arc_easy":null,"boolq":null,"hellaswag":null,"lambada_openai":null,"openbookqa":null,"piqa":null,"sciq":null,"winogrande":null,"harness":"lm-eval==0.4.12","num_fewshot":0,"seed":1234,"decontaminated":true,"scope":"general-domain","source":"harrrshall/BarunLM-35M benchmark_results.json (comparison table, macro only)"}
|