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": 76305.23101302811,
-      "required_pages_per_second_upper_estimate": 0.006613944030369367
+      "required_token_positions_per_second": 76133.77882186646,
+      "required_pages_per_second_upper_estimate": 0.006599082988981081
     },
     {
       "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": 9538.153876628514,
-      "required_pages_per_second_upper_estimate": 0.0038049121894959763
+      "required_token_positions_per_second": 9516.722352733308,
+      "required_pages_per_second_upper_estimate": 0.003796362834184342
     },
     {
       "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": 19076.307753257028,
-      "required_pages_per_second_upper_estimate": 0.0044129680671662124
+      "required_token_positions_per_second": 19033.444705466616,
+      "required_pages_per_second_upper_estimate": 0.004403052455423777
     },
     {
       "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": 28614.46162988554,
-      "required_pages_per_second_upper_estimate": 0.0017424705847823975
+      "required_token_positions_per_second": 28550.167058199924,
+      "required_pages_per_second_upper_estimate": 0.001738555382694291
     },
     {
       "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": 28614.46162988554,
-      "required_pages_per_second_upper_estimate": 0.0022028204661680394
+      "required_token_positions_per_second": 28550.167058199924,
+      "required_pages_per_second_upper_estimate": 0.0021978708920609154
     },
     {
       "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": 28614.46162988554,
-      "required_pages_per_second_upper_estimate": 0.35529584700057787
+      "required_token_positions_per_second": 28550.167058199924,
+      "required_pages_per_second_upper_estimate": 0.3544975236003318
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,17 +155,17 @@
   "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": 0.05702453298437682,
-    "producer_p95_seconds": 0.13808629599998312,
-    "gpu_update_seconds": 0.9662247138392376,
-    "producer_batches_per_second": 17.536311963729244,
-    "gpu_batches_per_second": 1.0349559327938924,
-    "producer_headroom": 16.944018008949886,
+    "producer_batch_seconds": 0.0540352513828104,
+    "producer_p95_seconds": 0.13203073000022414,
+    "gpu_update_seconds": 0.9684006382042933,
+    "producer_batches_per_second": 18.506437453497593,
+    "gpu_batches_per_second": 1.0326304636212358,
+    "producer_headroom": 17.921645840854907,
     "sustainable": true,
-    "buffer_seconds": 60.872156971871966,
+    "buffer_seconds": 60.040839568666186,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 64,

metrics/void-byte-v1/1790696019647648547-3fdea0a8/train.jsonl

@@ -11,3 +11,4 @@
 {"step": 276100, "loss": 0.6443329565227032, "lr": 0.0003, "grad_norm": 0.178448028994917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40412682300, "tokens_per_second": 152802.36093792898, "data_wait_s": 1.1008000001311302e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 276200, "loss": 0.6206479296088219, "lr": 0.0003, "grad_norm": 0.17355170529814581, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40427451945, "tokens_per_second": 152518.9564074561, "data_wait_s": 1.4293999811343383e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 276300, "loss": 0.5729329362511635, "lr": 0.0003, "grad_norm": 0.1990077812077371, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40442115845, "tokens_per_second": 150903.31363837697, "data_wait_s": 1.4895000276737846e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 276400, "loss": 0.5964973084628582, "lr": 0.0003, "grad_norm": 0.15778406929401198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40456654049, "tokens_per_second": 149641.8913269177, "data_wait_s": 1.4819000170973595e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790696019.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.22407.0.1790696019647648547-3fdea0a8

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7768359cdeb9de15b5421f31c3d42259fae3f6a63eecd7f274021f28c3094d59
-size 27699
+oid sha256:085239fba7380171c101ce49a012490c7c23882889f43cc13448548b6c39aa4a
+size 29817