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": 76228.00974719698,
-      "required_pages_per_second_upper_estimate": 0.006607250686762633
+      "required_token_positions_per_second": 74976.1524460553,
+      "required_pages_per_second_upper_estimate": 0.006498742868702995
     },
     {
       "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": 9528.501218399622,
-      "required_pages_per_second_upper_estimate": 0.00380106159980837
+      "required_token_positions_per_second": 9372.019055756913,
+      "required_pages_per_second_upper_estimate": 0.003738638525513369
     },
     {
       "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": 19057.002436799245,
-      "required_pages_per_second_upper_estimate": 0.004408502121965669
+      "required_token_positions_per_second": 18744.038111513826,
+      "required_pages_per_second_upper_estimate": 0.004336103333294892
     },
     {
       "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": 28585.503655198867,
-      "required_pages_per_second_upper_estimate": 0.0017407071960547304
+      "required_token_positions_per_second": 28116.057167270737,
+      "required_pages_per_second_upper_estimate": 0.0017121203679352792
     },
     {
       "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": 28585.503655198867,
-      "required_pages_per_second_upper_estimate": 0.002200591201115855
+      "required_token_positions_per_second": 28116.057167270737,
+      "required_pages_per_second_upper_estimate": 0.0021644519109640967
     },
     {
       "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": 28585.503655198867,
-      "required_pages_per_second_upper_estimate": 0.35493628587107623
+      "required_token_positions_per_second": 28116.057167270737,
+      "required_pages_per_second_upper_estimate": 0.3491073316273357
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11422511410142988,
-    "producer_p95_seconds": 0.3618045629991684,
-    "gpu_update_seconds": 0.9672035285259575,
-    "producer_batches_per_second": 8.754642163123757,
-    "gpu_batches_per_second": 1.033908552343709,
-    "producer_headroom": 8.467520791155419,
+    "producer_batch_seconds": 0.1403276078744966,
+    "producer_p95_seconds": 0.40225555100187194,
+    "gpu_update_seconds": 0.9833526740792236,
+    "producer_batches_per_second": 7.126181477377994,
+    "gpu_batches_per_second": 1.01692915101529,
+    "producer_headroom": 7.007549611753482,
     "sustainable": true,
-    "buffer_seconds": 60.93382229713532,
+    "buffer_seconds": 60.96786579291186,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354783.7727048669,
-    "upload_bytes_per_second": 5691532.768220302,
+    "generation_bytes_per_second": 382884.7909341969,
+    "upload_bytes_per_second": 5690769.518835166,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.12833867892746,
+    "estimated_full_shards_per_hour": 82.15816303271704,
     "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": 1.11581381400174,
-      "bytes": 5950902,
-      "pages": 37,
-      "shards": 2
-    },
-    {
-      "seconds": 0.9459429260023171,
-      "bytes": 2335478,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.2669872590049636,
       "bytes": 5719604,
@@ -562,6 +550,18 @@
       "bytes": 1021880,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 1.0568422009964706,
+      "bytes": 6256138,
+      "pages": 38,
+      "shards": 2
+    },
+    {
+      "seconds": 0.9614676109995344,
+      "bytes": 1727275,
+      "pages": 10,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -346,3 +346,4 @@
 {"step": 314800, "loss": 0.662660850211978, "lr": 0.0003, "grad_norm": 0.15615778817110823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46078226255, "tokens_per_second": 150534.09911022458, "data_wait_s": 1.2757998774759471e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 314900, "loss": 0.5819839462637901, "lr": 0.0003, "grad_norm": 0.13615964844314887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46092755066, "tokens_per_second": 149192.2055798279, "data_wait_s": 9.184994269162416e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 315000, "loss": 0.5823196284472942, "lr": 0.0003, "grad_norm": 0.1495047343906294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46107274235, "tokens_per_second": 148961.74686176443, "data_wait_s": 1.3094999303575605e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 315100, "loss": 0.5647122040390968, "lr": 0.0003, "grad_norm": 0.17350227340445543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46121906842, "tokens_per_second": 133042.40348086104, "data_wait_s": 9.980998584069312e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d9b00604a9dca494178d2f26dee5b472ba4370647550d72e553ca4c030de49f0
-size 742710
+oid sha256:06279147bbb384f30bf7548c91b76bd07b1c13d19793879501714aff2bc60a85
+size 744828