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": 75464.02981047281,
-      "required_pages_per_second_upper_estimate": 0.00654103084213683
+      "required_token_positions_per_second": 76042.15455356624,
+      "required_pages_per_second_upper_estimate": 0.006591141229624403
     },
     {
       "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": 9433.003726309102,
-      "required_pages_per_second_upper_estimate": 0.003762966222398716
+      "required_token_positions_per_second": 9505.26931919578,
+      "required_pages_per_second_upper_estimate": 0.0037917940478681108
     },
     {
       "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": 18866.007452618203,
-      "required_pages_per_second_upper_estimate": 0.004364318793772301
+      "required_token_positions_per_second": 19010.53863839156,
+      "required_pages_per_second_upper_estimate": 0.0043977535399389745
     },
     {
       "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": 28299.011178927307,
-      "required_pages_per_second_upper_estimate": 0.001723261307359649
+      "required_token_positions_per_second": 28515.80795758734,
+      "required_pages_per_second_upper_estimate": 0.0017364630937352539
     },
     {
       "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": 28299.011178927307,
-      "required_pages_per_second_upper_estimate": 0.002178536217230538
+      "required_token_positions_per_second": 28515.80795758734,
+      "required_pages_per_second_upper_estimate": 0.0021952258333836805
     },
     {
       "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": 28299.011178927307,
-      "required_pages_per_second_upper_estimate": 0.3513790081444219
+      "required_token_positions_per_second": 28515.80795758734,
+      "required_pages_per_second_upper_estimate": 0.35407089856323604
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.5098291875702614,
-    "producer_p95_seconds": 0.32673711700044805,
-    "gpu_update_seconds": 0.9769952676151428,
-    "producer_batches_per_second": 1.961441252051083,
-    "gpu_batches_per_second": 1.0235464112748591,
-    "producer_headroom": 1.9163188209590285,
+    "producer_batch_seconds": 0.08478709657796912,
+    "producer_p95_seconds": 0.2320374669980083,
+    "gpu_update_seconds": 0.9695674778397279,
+    "producer_batches_per_second": 11.794247478216368,
+    "gpu_batches_per_second": 1.0313877299474588,
+    "producer_headroom": 11.435318780471814,
     "sustainable": true,
-    "buffer_seconds": 60.573706592138855,
+    "buffer_seconds": 60.113183626063126,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 268194.45773466956,
-    "upload_bytes_per_second": 5270319.018307369,
+    "generation_bytes_per_second": 280924.80396095593,
+    "upload_bytes_per_second": 5329885.641236527,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.548287382412575,
+    "estimated_full_shards_per_hour": 60.2799233352805,
     "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": 3.174646825998934,
-      "bytes": 11736199,
-      "pages": 33,
-      "shards": 4
-    },
     {
       "seconds": 1.0644383409999136,
       "bytes": 5968582,
@@ -562,6 +556,12 @@
       "bytes": 5733304,
       "pages": 21,
       "shards": 2
+    },
+    {
+      "seconds": 1.3411155760004476,
+      "bytes": 6738025,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -195,3 +195,4 @@
 {"step": 174500, "loss": 0.6052178870886564, "lr": 0.0003, "grad_norm": 0.14196068976940465, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25539019678, "tokens_per_second": 148059.00687164412, "data_wait_s": 1.4252000255510211e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 174600, "loss": 0.6640870906412601, "lr": 0.0003, "grad_norm": 0.14839676530754553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25553572757, "tokens_per_second": 130426.6337239318, "data_wait_s": 1.220099875354208e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 174700, "loss": 0.6221353113651276, "lr": 0.0003, "grad_norm": 0.14574767237304562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25568218855, "tokens_per_second": 149726.34742037626, "data_wait_s": 1.2144999345764518e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 174800, "loss": 0.5843663346022367, "lr": 0.0003, "grad_norm": 0.1274422902860118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25582857314, "tokens_per_second": 149284.75482356458, "data_wait_s": 1.2662003427976742e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:66c67622dd46f3cca6d841c0c2e212fe7a8be1319e147c584084cba58e6da221
-size 419760
+oid sha256:c7e9643ad97d92c6bedb35304690d60fae77ee8a07cbe9e3336bd3df7e729602
+size 421878