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": 76763.5627019843,
-      "required_pages_per_second_upper_estimate": 0.005310681383728338
+      "required_token_positions_per_second": 76961.69158953922,
+      "required_pages_per_second_upper_estimate": 0.005324388399891745
     },
     {
       "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": 9595.445337748037,
-      "required_pages_per_second_upper_estimate": 0.004017412587230972
+      "required_token_positions_per_second": 9620.211448692402,
+      "required_pages_per_second_upper_estimate": 0.004027781640708172
     },
     {
       "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": 19190.890675496074,
-      "required_pages_per_second_upper_estimate": 0.003996956429885115
+      "required_token_positions_per_second": 19240.422897384804,
+      "required_pages_per_second_upper_estimate": 0.004007272685452004
     },
     {
       "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": 28786.336013244112,
-      "required_pages_per_second_upper_estimate": 0.0017256968898039225
+      "required_token_positions_per_second": 28860.634346077208,
+      "required_pages_per_second_upper_estimate": 0.0017301509614102812
     },
     {
       "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": 28786.336013244112,
-      "required_pages_per_second_upper_estimate": 0.0022380110073189794
+      "required_token_positions_per_second": 28860.634346077208,
+      "required_pages_per_second_upper_estimate": 0.0022437873758929246
     },
     {
       "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": 28786.336013244112,
-      "required_pages_per_second_upper_estimate": 0.36924021001839524
+      "required_token_positions_per_second": 28860.634346077208,
+      "required_pages_per_second_upper_estimate": 0.37019322925664383
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.12484423533607014,
-    "producer_p95_seconds": 0.2544459149976319,
-    "gpu_update_seconds": 0.9604556824209798,
-    "producer_batches_per_second": 8.009981376457507,
-    "gpu_batches_per_second": 1.0411724541827296,
-    "producer_headroom": 7.693232129104834,
+    "producer_batch_seconds": 0.08797849044543682,
+    "producer_p95_seconds": 0.23795810999945388,
+    "gpu_update_seconds": 0.9579831014268045,
+    "producer_batches_per_second": 11.366414619493701,
+    "gpu_batches_per_second": 1.043859749207075,
+    "producer_headroom": 10.88883312928555,
     "sustainable": true,
-    "buffer_seconds": 60.50870799252173,
+    "buffer_seconds": 60.352935389888685,
     "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": 257624.67175665637,
-    "upload_bytes_per_second": 3980381.792086119,
+    "generation_bytes_per_second": 269244.3843032545,
+    "upload_bytes_per_second": 3938075.587610757,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.280257363555606,
+    "estimated_full_shards_per_hour": 57.773577182991275,
     "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": 1.6360844549999456,
-      "bytes": 7454978,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.568186391999916,
       "bytes": 1164036,
@@ -562,6 +556,12 @@
       "bytes": 7345474,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.195227366002655,
+      "bytes": 1189304,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -194,3 +194,4 @@
 {"step": 819400, "loss": 0.5596259832382202, "lr": 3e-05, "grad_norm": 0.2890894522266533, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119937252394, "tokens_per_second": 152099.39793271886, "data_wait_s": 1.1822998203570023e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 819500, "loss": 0.5511067509651184, "lr": 3e-05, "grad_norm": 0.2806126738103894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119951999815, "tokens_per_second": 152030.24434130485, "data_wait_s": 9.433999366592616e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 819600, "loss": 0.5610944032669067, "lr": 3e-05, "grad_norm": 0.2856196797202541, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119966755033, "tokens_per_second": 153042.9228750421, "data_wait_s": 9.103001502808183e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 819700, "loss": 0.48045778274536133, "lr": 3e-05, "grad_norm": 0.2818042216552244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119981363123, "tokens_per_second": 143385.86027960118, "data_wait_s": 1.123100082622841e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ad371ad9b1d41d0b04dfd9b27aea380512205bf82bc05c8a24e1e654bead53d2
-size 417642
+oid sha256:9b038fa41f2b4e8ddb7134989d909415f318ed1b9ed3c1413dc892228d86fb94
+size 419760