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": 75454.68869670457,
-      "required_pages_per_second_upper_estimate": 0.0065402211780702345
+      "required_token_positions_per_second": 75467.24692020327,
+      "required_pages_per_second_upper_estimate": 0.006541309693054575
     },
     {
       "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": 9431.836087088072,
-      "required_pages_per_second_upper_estimate": 0.003762500433655685
+      "required_token_positions_per_second": 9433.405865025408,
+      "required_pages_per_second_upper_estimate": 0.0037631266415451606
     },
     {
       "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": 18863.672174176143,
-      "required_pages_per_second_upper_estimate": 0.004363778568204227
+      "required_token_positions_per_second": 18866.811730050817,
+      "required_pages_per_second_upper_estimate": 0.004364504849201537
     },
     {
       "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": 28295.508261264218,
-      "required_pages_per_second_upper_estimate": 0.0017230479980523555
+      "required_token_positions_per_second": 28300.217595076225,
+      "required_pages_per_second_upper_estimate": 0.0017233347717735418
     },
     {
       "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": 28295.508261264218,
-      "required_pages_per_second_upper_estimate": 0.0021782665529321366
+      "required_token_positions_per_second": 28300.217595076225,
+      "required_pages_per_second_upper_estimate": 0.002178629090485255
     },
     {
       "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": 28295.508261264218,
-      "required_pages_per_second_upper_estimate": 0.3513355136305576
+      "required_token_positions_per_second": 28300.217595076225,
+      "required_pages_per_second_upper_estimate": 0.35139398779537634
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10550399664049337,
-    "producer_p95_seconds": 0.2610298020008486,
-    "gpu_update_seconds": 0.9771162173414415,
-    "producer_batches_per_second": 9.478313920253814,
-    "gpu_batches_per_second": 1.0234197143107717,
-    "producer_headroom": 9.261414244533137,
+    "producer_batch_seconds": 0.05953606667193867,
+    "producer_p95_seconds": 0.1951913049997529,
+    "gpu_update_seconds": 0.9769536190707699,
+    "producer_batches_per_second": 16.79654125473716,
+    "gpu_batches_per_second": 1.0235900461182084,
+    "producer_headroom": 16.40944176668696,
     "sustainable": true,
-    "buffer_seconds": 60.58120547516937,
+    "buffer_seconds": 60.57112438238774,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 436899.68538193393,
-    "upload_bytes_per_second": 5948721.523509742,
+    "generation_bytes_per_second": 491022.5503726372,
+    "upload_bytes_per_second": 5846447.101302481,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.74850197881234,
+    "estimated_full_shards_per_hour": 105.36200889000261,
     "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": [
@@ -394,6 +394,12 @@
       "bytes": 7189609,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 0.8812644179997733,
+      "bytes": 1118965,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -90,3 +90,4 @@
 {"step": 94000, "loss": 0.6277541741728783, "lr": 0.0003, "grad_norm": 0.14381173255763774, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13756436345, "tokens_per_second": 149511.49512423552, "data_wait_s": 1.3134998880559579e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 94100, "loss": 0.5834298804402351, "lr": 0.0003, "grad_norm": 0.13784254449059272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13771106008, "tokens_per_second": 148703.5214312035, "data_wait_s": 1.162699845735915e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 94200, "loss": 0.6251903437077999, "lr": 0.0003, "grad_norm": 0.1315070496077938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13785653862, "tokens_per_second": 148819.15678524968, "data_wait_s": 6.8160006776452065e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 94300, "loss": 0.6212357766926289, "lr": 0.0003, "grad_norm": 0.13607407178389822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13800147761, "tokens_per_second": 147773.20663485298, "data_wait_s": 9.108000085689127e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e1a44019ac451e3a74a2b293db41ed9c96b4dc4a2d62c48be9c3fd1ae1df5e47
-size 195804
+oid sha256:2798d6541e156bdd85f0e7cdc8ad44668470a64ef9eddf5f61d407a0d531ed07
+size 197922