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": 75827.43369864502,
-      "required_pages_per_second_upper_estimate": 0.006572529770124863
+      "required_token_positions_per_second": 75773.78078331477,
+      "required_pages_per_second_upper_estimate": 0.006567879271406106
     },
     {
       "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": 9478.429212330628,
-      "required_pages_per_second_upper_estimate": 0.0037810871279442427
+      "required_token_positions_per_second": 9471.722597914346,
+      "required_pages_per_second_upper_estimate": 0.0037784117591807666
     },
     {
       "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": 18956.858424661255,
-      "required_pages_per_second_upper_estimate": 0.004385335567231963
+      "required_token_positions_per_second": 18943.44519582869,
+      "required_pages_per_second_upper_estimate": 0.004382232652806315
     },
     {
       "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": 28435.287636991885,
-      "required_pages_per_second_upper_estimate": 0.0017315598286684634
+      "required_token_positions_per_second": 28415.167793743036,
+      "required_pages_per_second_upper_estimate": 0.001730334635775268
     },
     {
       "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": 28435.287636991885,
-      "required_pages_per_second_upper_estimate": 0.0021890271562097293
+      "required_token_positions_per_second": 28415.167793743036,
+      "required_pages_per_second_upper_estimate": 0.002187478274981142
     },
     {
       "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": 28435.287636991885,
-      "required_pages_per_second_upper_estimate": 0.3530711056656181
+      "required_token_positions_per_second": 28415.167793743036,
+      "required_pages_per_second_upper_estimate": 0.3528212845492511
     }
   ],
   "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": 0.07278857979688524,
-    "producer_p95_seconds": 0.14494627700014462,
-    "gpu_update_seconds": 0.9723130060422638,
-    "producer_batches_per_second": 13.73841889470128,
-    "gpu_batches_per_second": 1.0284753919629588,
-    "producer_headroom": 13.358043373774835,
+    "producer_batch_seconds": 0.12474190689843212,
+    "producer_p95_seconds": 0.1442642329998307,
+    "gpu_update_seconds": 0.9730014688172292,
+    "producer_batches_per_second": 8.016552134434054,
+    "gpu_batches_per_second": 1.027747677725081,
+    "producer_headroom": 7.800117001654227,
     "sustainable": true,
-    "buffer_seconds": 60.283406374620355,
+    "buffer_seconds": 60.326091066668205,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
     "upload_pages_per_commit": 148,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 645582.3908888594,
+    "generation_bytes_per_second": 638677.8404645902,
     "upload_bytes_per_second": 2633196.0864612823,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.26347634792012,
+    "estimated_full_shards_per_hour": 68.52269845224991,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -15,3 +15,4 @@
 {"step": 1500, "loss": 0.916433073580265, "lr": 0.0003, "grad_norm": 0.2645044588396853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 219980853, "tokens_per_second": 148486.87014515663, "data_wait_s": 1.8279999949299963e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 1600, "loss": 0.8880256526172161, "lr": 0.0003, "grad_norm": 0.22562145193875374, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 234611375, "tokens_per_second": 149071.75160437857, "data_wait_s": 8.28099973659846e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 1700, "loss": 0.9110855236649513, "lr": 0.0003, "grad_norm": 0.2608335000330014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 249356749, "tokens_per_second": 150226.706145175, "data_wait_s": 1.4890999864292098e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 1800, "loss": 0.9287391342222691, "lr": 0.0003, "grad_norm": 0.23552261001673377, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 263963539, "tokens_per_second": 149206.06424722384, "data_wait_s": 1.4887000361341052e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:107de7f53dced94ff1569baa6aaa22b49dc9e3e07802ed1f876875578a82c4aa
-size 35537
+oid sha256:426833c87931dffc5afab24b0c5789a2f1670cf4c349e05621598e1a125824db
+size 37620