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": 74775.34032285692,
-      "required_pages_per_second_upper_estimate": 0.006481336982817832
+      "required_token_positions_per_second": 75475.06458312603,
+      "required_pages_per_second_upper_estimate": 0.006541987308263035
     },
     {
       "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": 9346.917540357115,
-      "required_pages_per_second_upper_estimate": 0.0037286251557192896
+      "required_token_positions_per_second": 9434.383072890754,
+      "required_pages_per_second_upper_estimate": 0.0037635164643732067
     },
     {
       "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": 18693.83508071423,
-      "required_pages_per_second_upper_estimate": 0.004324489745662571
+      "required_token_positions_per_second": 18868.76614578151,
+      "required_pages_per_second_upper_estimate": 0.00436495696888429
     },
     {
       "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": 28040.752621071344,
-      "required_pages_per_second_upper_estimate": 0.0017075347161641557
+      "required_token_positions_per_second": 28303.149218672264,
+      "required_pages_per_second_upper_estimate": 0.0017235132922695001
     },
     {
       "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": 28040.752621071344,
-      "required_pages_per_second_upper_estimate": 0.0021586547585413416
+      "required_token_positions_per_second": 28303.149218672264,
+      "required_pages_per_second_upper_estimate": 0.0021788547749849317
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28040.752621071344,
-      "required_pages_per_second_upper_estimate": 0.3481723011916429
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28303.149218672264,
+      "required_pages_per_second_upper_estimate": 0.3514303887489261
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.4045382722186446,
-    "producer_p95_seconds": 0.28634099799819523,
-    "gpu_update_seconds": 0.9859935064376193,
-    "producer_batches_per_second": 2.4719539995946804,
-    "gpu_batches_per_second": 1.0142054622783327,
-    "producer_headroom": 2.4373305918128563,
+    "producer_batch_seconds": 0.18298452964069156,
+    "producer_p95_seconds": 0.3307031780022953,
+    "gpu_update_seconds": 0.976852426787898,
+    "producer_batches_per_second": 5.464942866829235,
+    "gpu_batches_per_second": 1.0236960799577641,
+    "producer_headroom": 5.33844270171935,
     "sustainable": true,
-    "buffer_seconds": 60.14560389269478,
+    "buffer_seconds": 60.56485046084968,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318472.4201578741,
-    "upload_bytes_per_second": 5502684.02431877,
+    "generation_bytes_per_second": 326064.18071030086,
+    "upload_bytes_per_second": 5732622.116984269,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.33676770736854,
+    "estimated_full_shards_per_hour": 69.96578279477853,
     "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": 1.5069925719999446,
-      "bytes": 1775692,
-      "pages": 21,
-      "shards": 2
-    },
     {
       "seconds": 1.199250067000321,
       "bytes": 13436756,
@@ -562,6 +556,12 @@
       "bytes": 1100982,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 1.2673739959973318,
+      "bytes": 17927773,
+      "pages": 41,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -208,3 +208,4 @@
 {"step": 20800, "loss": 0.7289977855980396, "lr": 0.0003, "grad_norm": 0.15577767153490452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3043656574, "tokens_per_second": 148297.71208878103, "data_wait_s": 1.1731997801689431e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 20900, "loss": 0.6284867636859417, "lr": 0.0003, "grad_norm": 0.1758226288225565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3058158501, "tokens_per_second": 147687.78176371922, "data_wait_s": 8.006001735338941e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 21000, "loss": 0.6812669988721609, "lr": 0.0003, "grad_norm": 0.13401157085151533, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3072774075, "tokens_per_second": 149026.2730084698, "data_wait_s": 1.1579999409150332e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 21100, "loss": 0.6893913336098194, "lr": 0.0003, "grad_norm": 0.1397953381680224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3087368786, "tokens_per_second": 147324.48810078556, "data_wait_s": 1.7758997273631394e-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 sha256:59b1d5ff333e596df741ca25cb04b348d14ff033cfe49da89cd70f9475686cb1
-size 442303
+oid sha256:baabc1f6e96edb26e165ff9e0bd705596a45a3e81f5c26df8060307fcc26f13e
+size 444421