Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77310.12042603339,
-      "required_pages_per_second_upper_estimate": 0.0053484935152666975
+      "required_token_positions_per_second": 77184.15240120003,
+      "required_pages_per_second_upper_estimate": 0.005339778755022639
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9663.765053254174,
-      "required_pages_per_second_upper_estimate": 0.004046016625435499
+      "required_token_positions_per_second": 9648.019050150004,
+      "required_pages_per_second_upper_estimate": 0.004039424102749718
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19327.530106508348,
-      "required_pages_per_second_upper_estimate": 0.004025414819940856
+      "required_token_positions_per_second": 19296.038100300007,
+      "required_pages_per_second_upper_estimate": 0.004018855865547712
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28991.29515976252,
-      "required_pages_per_second_upper_estimate": 0.001737983877683205
+      "required_token_positions_per_second": 28944.05715045001,
+      "required_pages_per_second_upper_estimate": 0.001735152031153184
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28991.29515976252,
-      "required_pages_per_second_upper_estimate": 0.002253945679440776
+      "required_token_positions_per_second": 28944.05715045001,
+      "required_pages_per_second_upper_estimate": 0.0022502731319947724
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28991.29515976252,
-      "required_pages_per_second_upper_estimate": 0.37186920588194766
+      "required_token_positions_per_second": 28944.05715045001,
+      "required_pages_per_second_upper_estimate": 0.3712632874187095
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10889792547612842,
-    "producer_p95_seconds": 0.3357734739984153,
-    "gpu_update_seconds": 0.9536655691868882,
-    "producer_batches_per_second": 9.182911388144033,
-    "gpu_batches_per_second": 1.0485856177576143,
-    "producer_headroom": 8.757426415767137,
+    "producer_batch_seconds": 0.09236520344546761,
+    "producer_p95_seconds": 0.33202805299515603,
+    "gpu_update_seconds": 0.9552219944939592,
+    "producer_batches_per_second": 10.826587964919058,
+    "gpu_batches_per_second": 1.046877067073568,
+    "producer_headroom": 10.341794949414277,
     "sustainable": true,
-    "buffer_seconds": 60.080930858773954,
+    "buffer_seconds": 60.178985653119426,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 294257.68027223594,
-    "upload_bytes_per_second": 5083136.433436227,
+    "generation_bytes_per_second": 280175.74973057595,
+    "upload_bytes_per_second": 4969789.520127717,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.140848218205534,
+    "estimated_full_shards_per_hour": 60.11919373453101,
     "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": 1.4701623389992164,
-      "bytes": 17195357,
-      "pages": 39,
-      "shards": 5
-    },
     {
       "seconds": 0.8250811970028735,
       "bytes": 1111187,
@@ -562,6 +556,12 @@
       "bytes": 5214434,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.2110685700026806,
+      "bytes": 6209123,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -369,3 +369,4 @@
 {"step": 594000, "loss": 0.5627453327178955, "lr": 0.0003, "grad_norm": 0.16188618342335795, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86943668856, "tokens_per_second": 150438.6180611216, "data_wait_s": 1.107199932448566e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 594100, "loss": 0.5970128178596497, "lr": 0.0003, "grad_norm": 0.1625259251657857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86958180732, "tokens_per_second": 149836.70428047888, "data_wait_s": 1.5561003237962723e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 594200, "loss": 0.5892269015312195, "lr": 0.0003, "grad_norm": 0.17738979241311906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86972875993, "tokens_per_second": 153271.0979222932, "data_wait_s": 1.1397998605389148e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 594300, "loss": 0.5571348071098328, "lr": 0.0003, "grad_norm": 0.20933298242937576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86987617518, "tokens_per_second": 153495.50530834612, "data_wait_s": 1.3780001609120518e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9bc00a6575f621f21a68d1bef1a2fc21181cfac52e734edb09ae87c44c90949b
-size 791424
+oid sha256:d2546e32167024b68aca7be1820153fa889b3b9d2113fd03d4f6b8d9584357c4
+size 793542