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": 77279.17690851234,
-      "required_pages_per_second_upper_estimate": 0.005346352770925739
+      "required_token_positions_per_second": 76798.85547753122,
+      "required_pages_per_second_upper_estimate": 0.005313123019831197
     },
     {
       "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": 9659.897113564042,
-      "required_pages_per_second_upper_estimate": 0.004044397199859007
+      "required_token_positions_per_second": 9599.856934691403,
+      "required_pages_per_second_upper_estimate": 0.00401925963074654
     },
     {
       "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": 19319.794227128084,
-      "required_pages_per_second_upper_estimate": 0.004023803640274806
+      "required_token_positions_per_second": 19199.713869382806,
+      "required_pages_per_second_upper_estimate": 0.003998794068488451
     },
     {
       "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": 28979.691340692127,
-      "required_pages_per_second_upper_estimate": 0.0017372882464479406
+      "required_token_positions_per_second": 28799.57080407421,
+      "required_pages_per_second_upper_estimate": 0.0017264902953058314
     },
     {
       "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": 28979.691340692127,
-      "required_pages_per_second_upper_estimate": 0.002253043533547859
+      "required_token_positions_per_second": 28799.57080407421,
+      "required_pages_per_second_upper_estimate": 0.0022390399540923272
     },
     {
       "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": 28979.691340692127,
-      "required_pages_per_second_upper_estimate": 0.37172036455012286
+      "required_token_positions_per_second": 28799.57080407421,
+      "required_pages_per_second_upper_estimate": 0.3694099717047525
     }
   ],
   "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": 0.37539504663283196,
-    "producer_p95_seconds": 0.33063096499972744,
-    "gpu_update_seconds": 0.954047428420253,
-    "producer_batches_per_second": 2.6638604024471437,
-    "gpu_batches_per_second": 1.0481659194405428,
-    "producer_headroom": 2.5414491666252377,
+    "producer_batch_seconds": 0.1546972339138506,
+    "producer_p95_seconds": 0.42791969000245444,
+    "gpu_update_seconds": 0.9600143067440678,
+    "producer_batches_per_second": 6.464239693884187,
+    "gpu_batches_per_second": 1.0416511430871749,
+    "producer_headroom": 6.205762588351713,
     "sustainable": true,
-    "buffer_seconds": 60.10498799047594,
+    "buffer_seconds": 60.48090132487627,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 310614.38640406006,
-    "upload_bytes_per_second": 5210978.557360313,
+    "generation_bytes_per_second": 310104.5504380163,
+    "upload_bytes_per_second": 5215691.513909241,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.65061658946372,
+    "estimated_full_shards_per_hour": 66.5412176595246,
     "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": 1.106896608005627,
-      "bytes": 5997924,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.6541905339981895,
       "bytes": 16581039,
@@ -562,6 +556,12 @@
       "bytes": 10449012,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.3342199069957132,
+      "bytes": 7580590,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -415,3 +415,4 @@
 {"step": 438600, "loss": 0.6123660802841187, "lr": 0.0003, "grad_norm": 0.15547021280987575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64198395879, "tokens_per_second": 150770.17026214392, "data_wait_s": 1.3719996786676347e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 438700, "loss": 0.5546029210090637, "lr": 0.0003, "grad_norm": 0.15292847105504506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64212953981, "tokens_per_second": 151153.98076852626, "data_wait_s": 9.530005627311766e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 438800, "loss": 0.582893967628479, "lr": 0.0003, "grad_norm": 0.1656231395262307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64227561788, "tokens_per_second": 151095.56874685982, "data_wait_s": 1.3587996363639832e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 438900, "loss": 0.5672533512115479, "lr": 0.0003, "grad_norm": 0.15173435938648816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64242189125, "tokens_per_second": 151674.12938917405, "data_wait_s": 1.3265998859424144e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bfac1d1e53234002a936c4e3a56dfc5a108f14fb7b3c1b288cf74e778ed7dc55
-size 889635
+oid sha256:e1372517c2aa5c21712e80e61aec5f8d8a77908a669d4a15bc7b5a4e90fe74d3
+size 891753