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": 73976.36135483233,
-      "required_pages_per_second_upper_estimate": 0.006412083510863142
+      "required_token_positions_per_second": 74261.46789844203,
+      "required_pages_per_second_upper_estimate": 0.0064367958234673025
     },
     {
       "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": 9247.045169354042,
-      "required_pages_per_second_upper_estimate": 0.0036887845737011496
+      "required_token_positions_per_second": 9282.683487305254,
+      "required_pages_per_second_upper_estimate": 0.003703001231572225
     },
     {
       "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": 18494.090338708083,
-      "required_pages_per_second_upper_estimate": 0.004278282314986843
+      "required_token_positions_per_second": 18565.366974610508,
+      "required_pages_per_second_upper_estimate": 0.0042947709102769205
     },
     {
       "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": 27741.13550806213,
-      "required_pages_per_second_upper_estimate": 0.0016892896059513983
+      "required_token_positions_per_second": 27848.050461915762,
+      "required_pages_per_second_upper_estimate": 0.0016958001657016732
     },
     {
       "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": 27741.13550806213,
-      "required_pages_per_second_upper_estimate": 0.00213558940378864
+      "required_token_positions_per_second": 27848.050461915762,
+      "required_pages_per_second_upper_estimate": 0.002143820012895827
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27741.13550806213,
-      "required_pages_per_second_upper_estimate": 0.34445205940204043
+      "required_token_positions_per_second": 27848.050461915762,
+      "required_pages_per_second_upper_estimate": 0.34577958530756997
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.16920538428951204,
-    "producer_p95_seconds": 0.43409953200171003,
-    "gpu_update_seconds": 0.9966426930132308,
-    "producer_batches_per_second": 5.9099774170837875,
-    "gpu_batches_per_second": 1.0033686164663675,
-    "producer_headroom": 5.890135808609764,
+    "producer_batch_seconds": 0.10213931028931711,
+    "producer_p95_seconds": 0.3928964420047123,
+    "gpu_update_seconds": 0.9928163566713819,
+    "producer_batches_per_second": 9.790549761570022,
+    "gpu_batches_per_second": 1.007235621452393,
+    "producer_headroom": 9.720217944091814,
     "sustainable": true,
-    "buffer_seconds": 60.79520427380708,
+    "buffer_seconds": 60.5617977569543,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 264542.36934858205,
-    "upload_bytes_per_second": 6315561.677262525,
+    "generation_bytes_per_second": 299147.11639198085,
+    "upload_bytes_per_second": 6214279.952593561,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.76463423102471,
+    "estimated_full_shards_per_hour": 64.19000738925523,
     "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": 1.3802653160018963,
-      "bytes": 12066636,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 0.828995023999596,
       "bytes": 4743589,
@@ -562,6 +556,12 @@
       "bytes": 7783504,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.6935778650004067,
+      "bytes": 919630,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -404,3 +404,4 @@
 {"step": 80400, "loss": 0.589460164308548, "lr": 0.0003, "grad_norm": 0.13110316068270972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11765930227, "tokens_per_second": 147589.1460709096, "data_wait_s": 1.4668999938294291e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 80500, "loss": 0.6103362999856472, "lr": 0.0003, "grad_norm": 0.13600726218835835, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11780576634, "tokens_per_second": 149159.11343384674, "data_wait_s": 8.476999937556684e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 80600, "loss": 0.6215187516063452, "lr": 0.0003, "grad_norm": 0.13800425708100492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11795271071, "tokens_per_second": 149596.1864688307, "data_wait_s": 8.9899986051023e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 80700, "loss": 0.6559007130563259, "lr": 0.0003, "grad_norm": 0.14497893489856906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11810066328, "tokens_per_second": 149685.25251918632, "data_wait_s": 1.3229000614956021e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8ebc24a69ce203d162220df05879f8ab7825043e812cc284583bb0c0c28b0459
-size 866337
+oid sha256:98ff9c6557d69a1f15e93ad3bb10e9e46e471d7b484245fab257a2210fff2107
+size 868455