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": 75915.78547030968,
-      "required_pages_per_second_upper_estimate": 0.006580187877767237
+      "required_token_positions_per_second": 75097.33051527366,
+      "required_pages_per_second_upper_estimate": 0.006509246276619834
     },
     {
       "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": 9489.47318378871,
-      "required_pages_per_second_upper_estimate": 0.00378549273328096
+      "required_token_positions_per_second": 9387.166314409207,
+      "required_pages_per_second_upper_estimate": 0.003744680993461468
     },
     {
       "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": 18978.94636757742,
-      "required_pages_per_second_upper_estimate": 0.0043904452240911
+      "required_token_positions_per_second": 18774.332628818414,
+      "required_pages_per_second_upper_estimate": 0.004343111436708014
     },
     {
       "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": 28468.41955136613,
-      "required_pages_per_second_upper_estimate": 0.0017335773884241345
+      "required_token_positions_per_second": 28161.49894322762,
+      "required_pages_per_second_upper_estimate": 0.0017148875336764806
     },
     {
       "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": 28468.41955136613,
-      "required_pages_per_second_upper_estimate": 0.0021915777426932677
+      "required_token_positions_per_second": 28161.49894322762,
+      "required_pages_per_second_upper_estimate": 0.0021679501446681443
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28468.41955136613,
-      "required_pages_per_second_upper_estimate": 0.35348249315676183
+      "required_token_positions_per_second": 28161.49894322762,
+      "required_pages_per_second_upper_estimate": 0.3496715664008793
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.10231551803906669,
-    "producer_p95_seconds": 0.15708695500006797,
-    "gpu_update_seconds": 0.9711814156073599,
-    "producer_batches_per_second": 9.773688480159718,
-    "gpu_batches_per_second": 1.029673739560407,
-    "producer_headroom": 9.492024613866862,
+    "producer_batch_seconds": 0.15231465460158233,
+    "producer_p95_seconds": 0.16152967700008958,
+    "gpu_update_seconds": 0.9817659228912917,
+    "producer_batches_per_second": 6.565356449881687,
+    "gpu_batches_per_second": 1.01857273376836,
+    "producer_headroom": 6.445643234128389,
     "sustainable": true,
-    "buffer_seconds": 60.21324776765631,
+    "buffer_seconds": 60.86948721926009,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 169,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 681327.2791407491,
-    "upload_bytes_per_second": 3694044.83375718,
+    "generation_bytes_per_second": 378083.28968606365,
+    "upload_bytes_per_second": 4898646.269900984,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.09848680814197,
+    "estimated_full_shards_per_hour": 81.12787263809616,
     "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": [
@@ -202,6 +202,12 @@
       "bytes": 7790528,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.008963309000137,
+      "bytes": 10351006,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -29,3 +29,4 @@
 {"step": 357900, "loss": 0.5800473615527153, "lr": 0.0003, "grad_norm": 0.15822227173084832, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52386914088, "tokens_per_second": 149434.2085842559, "data_wait_s": 1.4267000096879201e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 358000, "loss": 0.580534003674984, "lr": 0.0003, "grad_norm": 0.1768261103372777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52401672123, "tokens_per_second": 150831.53612174874, "data_wait_s": 1.494899970566621e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 358100, "loss": 0.605397667735815, "lr": 0.0003, "grad_norm": 0.16282919517239758, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52416166657, "tokens_per_second": 128852.21877320574, "data_wait_s": 1.2104999768780544e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 358200, "loss": 0.5754309091717005, "lr": 0.0003, "grad_norm": 0.1601550239498113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52430698583, "tokens_per_second": 148090.7820259193, "data_wait_s": 1.5530000382568687e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1e2f5b18e7689aba5436da8469ed1f9d19f7cdacaf86a6ecf879f46f4376316e
-size 65823
+oid sha256:2cb60b441035595d0e162f4f7e45ebec78a2bea14cad2273621281cc81618540
+size 67941