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": 77589.50162974172,
-      "required_pages_per_second_upper_estimate": 0.005367821755192944
+      "required_token_positions_per_second": 77408.07338565747,
+      "required_pages_per_second_upper_estimate": 0.005355270128295671
     },
     {
       "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": 9698.687703717715,
-      "required_pages_per_second_upper_estimate": 0.004060638009916714
+      "required_token_positions_per_second": 9676.009173207183,
+      "required_pages_per_second_upper_estimate": 0.004051142982773525
     },
     {
       "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": 19397.37540743543,
-      "required_pages_per_second_upper_estimate": 0.0040399617541795165
+      "required_token_positions_per_second": 19352.018346414367,
+      "required_pages_per_second_upper_estimate": 0.0040305150745149655
     },
     {
       "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": 29096.063111153144,
-      "required_pages_per_second_upper_estimate": 0.0017442645563976737
+      "required_token_positions_per_second": 29028.02751962155,
+      "required_pages_per_second_upper_estimate": 0.001740185926569688
     },
     {
       "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": 29096.063111153144,
-      "required_pages_per_second_upper_estimate": 0.00226209092683588
+      "required_token_positions_per_second": 29028.02751962155,
+      "required_pages_per_second_upper_estimate": 0.0022568014588512397
     },
     {
       "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": 29096.063111153144,
-      "required_pages_per_second_upper_estimate": 0.3732130566713247
+      "required_token_positions_per_second": 29028.02751962155,
+      "required_pages_per_second_upper_estimate": 0.37234036915408414
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07868923936710814,
-    "producer_p95_seconds": 0.18843016000028,
-    "gpu_update_seconds": 0.9502316479854599,
-    "producer_batches_per_second": 12.708217896664495,
-    "gpu_batches_per_second": 1.0523749678513146,
-    "producer_headroom": 12.075750834905817,
+    "producer_batch_seconds": 0.08517447379692555,
+    "producer_p95_seconds": 0.20653366400074447,
+    "gpu_update_seconds": 0.952458791122176,
+    "producer_batches_per_second": 11.740606726661056,
+    "gpu_batches_per_second": 1.0499141898011266,
+    "producer_headroom": 11.182444089916478,
     "sustainable": true,
-    "buffer_seconds": 60.81482547106943,
+    "buffer_seconds": 60.00490384069709,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 168,
+    "upload_pages_per_commit": 166,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 609746.7482257115,
-    "upload_bytes_per_second": 5769395.051378128,
+    "generation_bytes_per_second": 605046.9348290886,
+    "upload_bytes_per_second": 5691370.183668633,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.41872899569755,
+    "estimated_full_shards_per_hour": 64.91449372126814,
     "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": 1.4687352659993849,
-      "bytes": 10700369,
-      "pages": 42,
-      "shards": 3
-    },
     {
       "seconds": 1.1858354729993152,
       "bytes": 2742758,
@@ -562,6 +556,12 @@
       "bytes": 10716254,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.0711626960001013,
+      "bytes": 2631436,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -120,3 +120,4 @@
 {"step": 537000, "loss": 0.6134182214736938, "lr": 0.0003, "grad_norm": 0.16458220191705777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78599136816, "tokens_per_second": 154497.98174207684, "data_wait_s": 1.591399995959364e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 537100, "loss": 0.5655831098556519, "lr": 0.0003, "grad_norm": 0.15928452333374893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78613716460, "tokens_per_second": 152636.14300174918, "data_wait_s": 1.2892998711322434e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 537200, "loss": 0.5576798915863037, "lr": 0.0003, "grad_norm": 0.18674772998216113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78628258807, "tokens_per_second": 152071.56662390172, "data_wait_s": 1.8783999621518888e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 537300, "loss": 0.5167000889778137, "lr": 0.0003, "grad_norm": 0.1550515354654404, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78643011973, "tokens_per_second": 154457.0379481357, "data_wait_s": 1.0593999832053669e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e5058108b4ef1a2042c7133ff5a5d662cab907ff6bab2fd2b51af1b46bef65b7
-size 260127
+oid sha256:0ec9c1167ec81f3e9b633284d9bcd76b80b84fcb413e19de6c34f99c9b1cbe10
+size 262245