Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75586.64812038859,
-      "required_pages_per_second_upper_estimate": 0.006551659086467064
+      "required_token_positions_per_second": 75784.46299527492,
+      "required_pages_per_second_upper_estimate": 0.006568805178464994
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9448.331015048574,
-      "required_pages_per_second_upper_estimate": 0.00376908050704028
+      "required_token_positions_per_second": 9473.057874409365,
+      "required_pages_per_second_upper_estimate": 0.0037789444209387927
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18896.662030097148,
-      "required_pages_per_second_upper_estimate": 0.004371410190769918
+      "required_token_positions_per_second": 18946.11574881873,
+      "required_pages_per_second_upper_estimate": 0.004382850438240432
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28344.993045145726,
-      "required_pages_per_second_upper_estimate": 0.0017260613617641433
+      "required_token_positions_per_second": 28419.173623228096,
+      "required_pages_per_second_upper_estimate": 0.001730578569773945
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28344.993045145726,
-      "required_pages_per_second_upper_estimate": 0.002182076028577984
+      "required_token_positions_per_second": 28419.173623228096,
+      "required_pages_per_second_upper_estimate": 0.0021877866548237475
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28344.993045145726,
-      "required_pages_per_second_upper_estimate": 0.351949949031448
+      "required_token_positions_per_second": 28419.173623228096,
+      "required_pages_per_second_upper_estimate": 0.3528710235448067
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.0517991353124998,
-    "producer_p95_seconds": 0.17165326200029085,
-    "gpu_update_seconds": 0.9754103645735384,
-    "producer_batches_per_second": 19.3053415653965,
-    "gpu_batches_per_second": 1.025209528542597,
-    "producer_headroom": 18.830630254520084,
+    "producer_batch_seconds": 0.06563927683594528,
+    "producer_p95_seconds": 0.16056643599995368,
+    "gpu_update_seconds": 0.9728643192285582,
+    "producer_batches_per_second": 15.234780884307085,
+    "gpu_batches_per_second": 1.027892564497544,
+    "producer_headroom": 14.821374733607664,
     "sustainable": true,
-    "buffer_seconds": 60.47544260355938,
+    "buffer_seconds": 60.317587792170606,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 503629.28425549454,
+    "generation_bytes_per_second": 529725.6631296694,
     "upload_bytes_per_second": 4537311.3303474495,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.06712051151874,
+    "estimated_full_shards_per_hour": 113.66679592530785,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -30,3 +30,4 @@
 {"step": 240100, "loss": 0.5598774999380112, "lr": 0.0003, "grad_norm": 0.15279531327039617, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35143001793, "tokens_per_second": 130649.21440094686, "data_wait_s": 9.452000085730106e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 240200, "loss": 0.6031834371387959, "lr": 0.0003, "grad_norm": 0.148073282968182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35157816049, "tokens_per_second": 151604.359841389, "data_wait_s": 1.8180999632022576e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 240300, "loss": 0.6386078968644142, "lr": 0.0003, "grad_norm": 0.14710874828634884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35172619114, "tokens_per_second": 150686.73625573123, "data_wait_s": 1.1144999916723464e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 240400, "loss": 0.5970120839774609, "lr": 0.0003, "grad_norm": 0.15777480089589546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35187296028, "tokens_per_second": 149482.5011255308, "data_wait_s": 1.2068999694747617e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d93f7d6432f32c6775030cd72e9f4876e45d720fb9a8e353679d65ff3d231991
-size 68724
+oid sha256:45cd010451aa947128f20a8a0b80095c6f617946e14e19673487f3177d7140a6
+size 70842