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": 75711.81142358141,
-      "required_pages_per_second_upper_estimate": 0.006562507924364326
+      "required_token_positions_per_second": 75411.24756335565,
+      "required_pages_per_second_upper_estimate": 0.006536455810732096
     },
     {
       "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": 9463.976427947677,
-      "required_pages_per_second_upper_estimate": 0.0037753216961655006
+      "required_token_positions_per_second": 9426.405945419456,
+      "required_pages_per_second_upper_estimate": 0.003760334268956221
     },
     {
       "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": 18927.952855895353,
-      "required_pages_per_second_upper_estimate": 0.004378648772618607
+      "required_token_positions_per_second": 18852.81189083891,
+      "required_pages_per_second_upper_estimate": 0.004361266232788617
     },
     {
       "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": 28391.92928384303,
-      "required_pages_per_second_upper_estimate": 0.0017289195324453964
+      "required_token_positions_per_second": 28279.217836258365,
+      "required_pages_per_second_upper_estimate": 0.0017220559966387536
     },
     {
       "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": 28391.92928384303,
-      "required_pages_per_second_upper_estimate": 0.0021856893101607267
+      "required_token_positions_per_second": 28279.217836258365,
+      "required_pages_per_second_upper_estimate": 0.0021770124709204032
     },
     {
       "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": 28391.92928384303,
-      "required_pages_per_second_upper_estimate": 0.3525327400305826
+      "required_token_positions_per_second": 28279.217836258365,
+      "required_pages_per_second_upper_estimate": 0.3511332410725302
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.17587550978120703,
-    "producer_p95_seconds": 0.2744712950006942,
-    "gpu_update_seconds": 0.9737978607791765,
-    "producer_batches_per_second": 5.685839951474891,
-    "gpu_batches_per_second": 1.0269071644908503,
-    "producer_headroom": 5.536858781479026,
+    "producer_batch_seconds": 0.0996449874139671,
+    "producer_p95_seconds": 0.28200468000068213,
+    "gpu_update_seconds": 0.9776790914121731,
+    "producer_batches_per_second": 10.035627741569982,
+    "gpu_batches_per_second": 1.0228305062304097,
+    "producer_headroom": 9.81162341212894,
     "sustainable": true,
-    "buffer_seconds": 60.37546736830895,
+    "buffer_seconds": 60.61610366755473,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 406922.95618581603,
-    "upload_bytes_per_second": 5292793.11060839,
+    "generation_bytes_per_second": 406074.91798424924,
+    "upload_bytes_per_second": 5343041.452071677,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.31619371586667,
+    "estimated_full_shards_per_hour": 87.13422445912941,
     "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": [
-    {
-      "seconds": 0.9120372190000126,
-      "bytes": 2656908,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.6420339440001044,
       "bytes": 5755818,
@@ -562,6 +556,12 @@
       "bytes": 11373221,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.0926308529997186,
+      "bytes": 7644433,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -123,3 +123,4 @@
 {"step": 167300, "loss": 0.6270021367818117, "lr": 0.0003, "grad_norm": 0.12998750910320842, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24485168322, "tokens_per_second": 149277.00870477536, "data_wait_s": 1.4106999515206553e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 167400, "loss": 0.5898058451712132, "lr": 0.0003, "grad_norm": 0.13277469101355197, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24499680534, "tokens_per_second": 148153.1811235543, "data_wait_s": 1.4287999874795787e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 167500, "loss": 0.6193031966686249, "lr": 0.0003, "grad_norm": 0.1399431171414283, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24514374012, "tokens_per_second": 150008.65919454175, "data_wait_s": 1.1673000699374825e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 167600, "loss": 0.57669323310256, "lr": 0.0003, "grad_norm": 0.12682190382748396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24528941816, "tokens_per_second": 149596.6484980393, "data_wait_s": 1.9020999388885684e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c64a2b3151c5a87e1f0493acc7de1026827097dd0ceb932d7d4437fb56f56d1
-size 266481
+oid sha256:48b3d43a92736e102fce4d95fa2f6627c5a682d48e1fd99abfe8951c2489f6f3
+size 268599