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": 76944.02837417864,
-      "required_pages_per_second_upper_estimate": 0.0053231664176687935
+      "required_token_positions_per_second": 77253.57605828736,
+      "required_pages_per_second_upper_estimate": 0.005344581644705018
     },
     {
       "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": 9618.00354677233,
-      "required_pages_per_second_upper_estimate": 0.004026857238280473
+      "required_token_positions_per_second": 9656.69700728592,
+      "required_pages_per_second_upper_estimate": 0.004043057382186175
     },
     {
       "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": 19236.00709354466,
-      "required_pages_per_second_upper_estimate": 0.004006352989964681
+      "required_token_positions_per_second": 19313.39401457184,
+      "required_pages_per_second_upper_estimate": 0.004022470644784296
     },
     {
       "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": 28854.01064031699,
-      "required_pages_per_second_upper_estimate": 0.001729753880363769
+      "required_token_positions_per_second": 28970.091021857763,
+      "required_pages_per_second_upper_estimate": 0.0017367127219926586
     },
     {
       "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": 28854.01064031699,
-      "required_pages_per_second_upper_estimate": 0.00224327241190467
+      "required_token_positions_per_second": 28970.091021857763,
+      "required_pages_per_second_upper_estimate": 0.002252297150985827
     },
     {
       "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": 28854.01064031699,
-      "required_pages_per_second_upper_estimate": 0.37010826747113285
+      "required_token_positions_per_second": 28970.091021857763,
+      "required_pages_per_second_upper_estimate": 0.3715972219681349
     }
   ],
   "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": 0.6677291778830181,
-    "producer_p95_seconds": 0.33473726399824955,
-    "gpu_update_seconds": 0.9582030153329236,
-    "producer_batches_per_second": 1.4976131538394353,
-    "gpu_batches_per_second": 1.0436201765160948,
-    "producer_headroom": 1.4350174398111963,
+    "producer_batch_seconds": 0.0927964930628491,
+    "producer_p95_seconds": 0.31433029499748955,
+    "gpu_update_seconds": 0.9543635875751908,
+    "producer_batches_per_second": 10.776269307103247,
+    "gpu_batches_per_second": 1.0478186856864065,
+    "producer_headroom": 10.28447903660347,
     "sustainable": true,
-    "buffer_seconds": 60.366789965974185,
+    "buffer_seconds": 60.12490601723702,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 336651.8775897813,
-    "upload_bytes_per_second": 4111056.838902051,
+    "generation_bytes_per_second": 336943.74593272776,
+    "upload_bytes_per_second": 4114063.170992691,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.23765607614592,
+    "estimated_full_shards_per_hour": 72.30028422819495,
     "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": 1.9298742110004241,
-      "bytes": 11871053,
-      "pages": 33,
-      "shards": 4
-    },
     {
       "seconds": 1.752713240999583,
       "bytes": 1247797,
@@ -562,6 +556,12 @@
       "bytes": 2384316,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 2.864481454998895,
+      "bytes": 16026880,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -334,3 +334,4 @@
 {"step": 875900, "loss": 0.5297982692718506, "lr": 3e-05, "grad_norm": 0.31422191003205474, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128207114873, "tokens_per_second": 151192.09840029027, "data_wait_s": 1.1267999070696533e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 876000, "loss": 0.532098650932312, "lr": 3e-05, "grad_norm": 0.2955324069768495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128221751007, "tokens_per_second": 151555.89544821964, "data_wait_s": 2.1441999706439674e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 876100, "loss": 0.5254486203193665, "lr": 3e-05, "grad_norm": 0.30969942609753287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128236336868, "tokens_per_second": 151478.03414473226, "data_wait_s": 1.1274001735728234e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 876200, "loss": 0.5402370691299438, "lr": 3e-05, "grad_norm": 0.3122375962551669, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128250930986, "tokens_per_second": 151039.32049315082, "data_wait_s": 1.5830002666916698e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:862ad3c892a757db51eb0904f7c66a5655cc324a9cf07dba210c123b4a5d3bd1
-size 717294
+oid sha256:6d21d786f6df3364d90b22637eed07f39389a10daa83f90af9b497ee4df82a6b
+size 719412