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": 76265.9403966976,
-      "required_pages_per_second_upper_estimate": 0.006610538419326994
+      "required_token_positions_per_second": 75977.4380536602,
+      "required_pages_per_second_upper_estimate": 0.006585531767435016
     },
     {
       "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": 9533.2425495872,
-      "required_pages_per_second_upper_estimate": 0.003802952987708313
+      "required_token_positions_per_second": 9497.179756707525,
+      "required_pages_per_second_upper_estimate": 0.003788567000441808
     },
     {
       "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": 19066.4850991744,
-      "required_pages_per_second_upper_estimate": 0.004410695769017004
+      "required_token_positions_per_second": 18994.35951341505,
+      "required_pages_per_second_upper_estimate": 0.004394010784118533
     },
     {
       "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": 28599.7276487616,
-      "required_pages_per_second_upper_estimate": 0.0017415733626351742
+      "required_token_positions_per_second": 28491.539270122576,
+      "required_pages_per_second_upper_estimate": 0.0017349852579966122
     },
     {
       "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": 28599.7276487616,
-      "required_pages_per_second_upper_estimate": 0.0022016862035148486
+      "required_token_positions_per_second": 28491.539270122576,
+      "required_pages_per_second_upper_estimate": 0.0021933575626426165
     },
     {
       "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": 28599.7276487616,
-      "required_pages_per_second_upper_estimate": 0.35511290026648124
+      "required_token_positions_per_second": 28491.539270122576,
+      "required_pages_per_second_upper_estimate": 0.3537695626869957
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "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": 0.0443331334922199,
-    "producer_p95_seconds": 0.12446291600008408,
-    "gpu_update_seconds": 0.9667224925897918,
-    "producer_batches_per_second": 22.556492655216708,
-    "gpu_batches_per_second": 1.0344230197034723,
-    "producer_headroom": 21.805868803734427,
+    "producer_batch_seconds": 0.04702988090627258,
+    "producer_p95_seconds": 0.1318802269997832,
+    "gpu_update_seconds": 0.9703933416118677,
+    "producer_batches_per_second": 21.26307744629278,
+    "gpu_batches_per_second": 1.0305099562399658,
+    "producer_headroom": 20.633548776059992,
     "sustainable": true,
-    "buffer_seconds": 60.90351703315689,
+    "buffer_seconds": 60.1643871799358,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 68,
+    "upload_pages_per_commit": 69,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -13,3 +13,4 @@
 {"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}
 {"step": 276500, "loss": 0.6285677775740623, "lr": 0.0003, "grad_norm": 0.1593182684767307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40471107394, "tokens_per_second": 149054.53261513283, "data_wait_s": 1.1183999959030189e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 276600, "loss": 0.5801319926977158, "lr": 0.0003, "grad_norm": 0.14238914330006738, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40485662186, "tokens_per_second": 149788.11770255343, "data_wait_s": 1.7440000192436855e-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:d3551df1198f9d58a8065e6dde306651b8346ed74adfc78838dd911581cccb32
-size 31935
+oid sha256:17fbc9407adc8cb7e7d775fd9db884f943c58db4f0bb285edef734b24d24f5c1
+size 34053