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": 74874.52467222826,
-      "required_pages_per_second_upper_estimate": 0.006489934030840905
+      "required_token_positions_per_second": 74546.89354989669,
+      "required_pages_per_second_upper_estimate": 0.0064615357955306926
     },
     {
       "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": 9359.315584028533,
-      "required_pages_per_second_upper_estimate": 0.00373357092070709
+      "required_token_positions_per_second": 9318.361693737086,
+      "required_pages_per_second_upper_estimate": 0.003717233801554606
     },
     {
       "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": 18718.631168057065,
-      "required_pages_per_second_upper_estimate": 0.004330225884073103
+      "required_token_positions_per_second": 18636.723387474172,
+      "required_pages_per_second_upper_estimate": 0.004311277960562955
     },
     {
       "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": 28077.946752085598,
-      "required_pages_per_second_upper_estimate": 0.0017097996436003997
+      "required_token_positions_per_second": 27955.08508121126,
+      "required_pages_per_second_upper_estimate": 0.0017023180124495219
     },
     {
       "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": 28077.946752085598,
-      "required_pages_per_second_upper_estimate": 0.0021615180657067632
+      "required_token_positions_per_second": 27955.08508121126,
+      "required_pages_per_second_upper_estimate": 0.002152059834179984
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28077.946752085598,
-      "required_pages_per_second_upper_estimate": 0.34863412781809106
+      "required_token_positions_per_second": 27955.08508121126,
+      "required_pages_per_second_upper_estimate": 0.347108597057393
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3070814999605318,
-    "producer_p95_seconds": 0.391260238000541,
-    "gpu_update_seconds": 0.9846873863006502,
-    "producer_batches_per_second": 3.256464489487406,
-    "gpu_batches_per_second": 1.015550736114207,
-    "producer_headroom": 3.2065995067342348,
+    "producer_batch_seconds": 0.46083899110163884,
+    "producer_p95_seconds": 0.3825809999980265,
+    "gpu_update_seconds": 0.9890150546736253,
+    "producer_batches_per_second": 2.169955275723291,
+    "gpu_batches_per_second": 1.0111069546155693,
+    "producer_headroom": 2.146118435658792,
     "sustainable": true,
-    "buffer_seconds": 60.065930564339666,
+    "buffer_seconds": 60.32991833509114,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 428661.99792522483,
-    "upload_bytes_per_second": 4281167.185560564,
+    "generation_bytes_per_second": 367591.4107940038,
+    "upload_bytes_per_second": 4234795.073533338,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.98088601415213,
+    "estimated_full_shards_per_hour": 78.87655966630064,
     "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": 1.1968161780023365,
-      "bytes": 11473828,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 0.9470581830028095,
       "bytes": 5899775,
@@ -562,6 +556,12 @@
       "bytes": 6743320,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.0128854309950839,
+      "bytes": 6059900,
+      "pages": 26,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -342,3 +342,4 @@
 {"step": 34200, "loss": 0.6966866441071033, "lr": 0.0003, "grad_norm": 0.14291184913415805, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5004702933, "tokens_per_second": 149746.034129097, "data_wait_s": 1.497899938840419e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 34300, "loss": 0.5933354459702969, "lr": 0.0003, "grad_norm": 0.15678194327714132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5019240778, "tokens_per_second": 147608.11027876544, "data_wait_s": 1.2222000805195421e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 34400, "loss": 0.6432075276970863, "lr": 0.0003, "grad_norm": 0.13552194736432954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5033696260, "tokens_per_second": 146893.72778329093, "data_wait_s": 1.0436997399665415e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 34500, "loss": 0.6013185773044825, "lr": 0.0003, "grad_norm": 0.13190112491699676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5048308953, "tokens_per_second": 148703.76954503203, "data_wait_s": 1.3197000953368843e-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 sha256:8e58b38b976057f19ae560340c2667b674ab884dc9df273ec065bf640dfab920
-size 727681
+oid sha256:0df4890da2b2476517a6439d44b26a3f89d58b56445741cae59285418f340d48
+size 729799