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": 76578.32374219413,
-      "required_pages_per_second_upper_estimate": 0.0052978661226764515
+      "required_token_positions_per_second": 76882.05646829403,
+      "required_pages_per_second_upper_estimate": 0.005318879057425064
     },
     {
       "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": 9572.290467774266,
-      "required_pages_per_second_upper_estimate": 0.0040077181267016235
+      "required_token_positions_per_second": 9610.257058536754,
+      "required_pages_per_second_upper_estimate": 0.004023613945421306
     },
     {
       "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": 19144.580935548533,
-      "required_pages_per_second_upper_estimate": 0.0039873113323240115
+      "required_token_positions_per_second": 19220.514117073508,
+      "required_pages_per_second_upper_estimate": 0.004003126211543005
     },
     {
       "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": 28716.871403322795,
-      "required_pages_per_second_upper_estimate": 0.0017215325925054575
+      "required_token_positions_per_second": 28830.77117561026,
+      "required_pages_per_second_upper_estimate": 0.0017283607099392087
     },
     {
       "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": 28716.871403322795,
-      "required_pages_per_second_upper_estimate": 0.0022326104394401257
+      "required_token_positions_per_second": 28830.77117561026,
+      "required_pages_per_second_upper_estimate": 0.0022414656457433245
     },
     {
       "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": 28716.871403322795,
-      "required_pages_per_second_upper_estimate": 0.3683491925876117
+      "required_token_positions_per_second": 28830.77117561026,
+      "required_pages_per_second_upper_estimate": 0.36981017657046805
     }
   ],
   "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": 0.21732652640628203,
-    "producer_p95_seconds": 0.25083491999976104,
-    "gpu_update_seconds": 0.9627789744812132,
-    "producer_batches_per_second": 4.601371109804358,
-    "gpu_batches_per_second": 1.0386599899928675,
-    "producer_headroom": 4.430103358304922,
+    "producer_batch_seconds": 0.06636041187508113,
+    "producer_p95_seconds": 0.23615663099917583,
+    "gpu_update_seconds": 0.9589753888855099,
+    "producer_batches_per_second": 15.069225336973957,
+    "gpu_batches_per_second": 1.0427796287474775,
+    "producer_headroom": 14.45101622772798,
     "sustainable": true,
-    "buffer_seconds": 60.655075392316434,
+    "buffer_seconds": 60.41544949978712,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 413662.09891703737,
-    "upload_bytes_per_second": 5444585.286216854,
+    "generation_bytes_per_second": 488680.52732284943,
+    "upload_bytes_per_second": 5471227.043594741,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.76225686677304,
+    "estimated_full_shards_per_hour": 104.85946526302445,
     "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": 0.842472636999446,
-      "bytes": 857193,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.0425618379995285,
       "bytes": 11991809,
@@ -562,6 +556,12 @@
       "bytes": 5962058,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2595222179988923,
+      "bytes": 5231936,
+      "pages": 41,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -223,3 +223,4 @@
 {"step": 547300, "loss": 0.558578610420227, "lr": 0.0003, "grad_norm": 0.1558924638887807, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80107842033, "tokens_per_second": 152632.96103963844, "data_wait_s": 1.0764000762719661e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 547400, "loss": 0.552259087562561, "lr": 0.0003, "grad_norm": 0.16428853215458364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80122289261, "tokens_per_second": 150614.76439718224, "data_wait_s": 1.2645999959204346e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 547500, "loss": 0.5929126143455505, "lr": 0.0003, "grad_norm": 0.16622130039256622, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80136905555, "tokens_per_second": 152401.2909801345, "data_wait_s": 9.955001587513834e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 547600, "loss": 0.5590025186538696, "lr": 0.0003, "grad_norm": 0.14889170783449715, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80151455347, "tokens_per_second": 152024.15229491683, "data_wait_s": 1.1046002327930182e-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 sha256:9704e1538c41d135997b4a634db6d7d532d1d0a93cf86ecbc765b4581a85ac3d
-size 479847
+oid sha256:f152f222a7d7a59d15567230ad1a682338c23a2dc2e225cf3364cff5b79c2b55
+size 481965