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": 76409.91123351322,
-      "required_pages_per_second_upper_estimate": 0.006623017446571423
+      "required_token_positions_per_second": 76147.15618754215,
+      "required_pages_per_second_upper_estimate": 0.006600242505133225
     },
     {
       "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": 9551.238904189153,
-      "required_pages_per_second_upper_estimate": 0.0038101320026285115
+      "required_token_positions_per_second": 9518.394523442768,
+      "required_pages_per_second_upper_estimate": 0.003797029888081526
     },
     {
       "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": 19102.477808378306,
-      "required_pages_per_second_upper_estimate": 0.004419022048841281
+      "required_token_positions_per_second": 19036.789046885537,
+      "required_pages_per_second_upper_estimate": 0.0044038261099526455
     },
     {
       "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": 28653.71671256746,
-      "required_pages_per_second_upper_estimate": 0.0017448610133622246
+      "required_token_positions_per_second": 28555.183570328307,
+      "required_pages_per_second_upper_estimate": 0.0017388608619633046
     },
     {
       "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": 28653.71671256746,
-      "required_pages_per_second_upper_estimate": 0.0022058424310979173
+      "required_token_positions_per_second": 28555.183570328307,
+      "required_pages_per_second_upper_estimate": 0.0021982570770511527
     },
     {
       "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": 28653.71671256746,
-      "required_pages_per_second_upper_estimate": 0.3557832637491769
+      "required_token_positions_per_second": 28555.183570328307,
+      "required_pages_per_second_upper_estimate": 0.35455981189178026
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.05647223249994937,
-    "producer_p95_seconds": 0.16571462199954112,
-    "gpu_update_seconds": 0.964901003152364,
-    "producer_batches_per_second": 17.707817731500107,
-    "gpu_batches_per_second": 1.0363757491524688,
-    "producer_headroom": 17.08629109276367,
+    "producer_batch_seconds": 0.054733491554650016,
+    "producer_p95_seconds": 0.1617851429991788,
+    "gpu_update_seconds": 0.9682305117004759,
+    "producer_batches_per_second": 18.27034913351956,
+    "gpu_batches_per_second": 1.0328119057555087,
+    "producer_headroom": 17.68990949049399,
     "sustainable": true,
-    "buffer_seconds": 60.788763198598936,
+    "buffer_seconds": 60.03029172542951,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -39,3 +39,4 @@
 {"step": 118900, "loss": 0.5925244968384504, "lr": 0.0003, "grad_norm": 0.1351402662102939, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17399398353, "tokens_per_second": 148453.03547389834, "data_wait_s": 1.2634000086109154e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 119000, "loss": 0.6057021394371986, "lr": 0.0003, "grad_norm": 0.15349554496346607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17414016155, "tokens_per_second": 150603.53231536408, "data_wait_s": 1.1053999514842872e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 119100, "loss": 0.5708722695708275, "lr": 0.0003, "grad_norm": 0.13491756279371664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17428635994, "tokens_per_second": 150197.70259587394, "data_wait_s": 1.6369000150007196e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 119200, "loss": 0.5866619851440191, "lr": 0.0003, "grad_norm": 0.138261813151199, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17443251462, "tokens_per_second": 150610.0283311074, "data_wait_s": 1.39579997266992e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7424eda9d744e7e1630694ecc4687f889dc3a031175560fbc022f980cefa3726
-size 87003
+oid sha256:2b5c8f45fca348e08decba9f1dd164065637fcd99a046d07f0e103bb883ad635
+size 89121