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": 75363.50435665406,
-      "required_pages_per_second_upper_estimate": 0.005213822098815464
+      "required_token_positions_per_second": 76481.10064421553,
+      "required_pages_per_second_upper_estimate": 0.005291140003169625
     },
     {
       "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": 9420.438044581757,
-      "required_pages_per_second_upper_estimate": 0.003944140688150107
+      "required_token_positions_per_second": 9560.137580526942,
+      "required_pages_per_second_upper_estimate": 0.004002629966592313
     },
     {
       "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": 18840.876089163514,
-      "required_pages_per_second_upper_estimate": 0.003924057622057409
+      "required_token_positions_per_second": 19120.275161053883,
+      "required_pages_per_second_upper_estimate": 0.003982249080483154
     },
     {
       "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": 28261.31413374527,
-      "required_pages_per_second_upper_estimate": 0.0016942226297899591
+      "required_token_positions_per_second": 28680.412741580825,
+      "required_pages_per_second_upper_estimate": 0.0017193469514031794
     },
     {
       "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": 28261.31413374527,
-      "required_pages_per_second_upper_estimate": 0.0021971928655151354
+      "required_token_positions_per_second": 28680.412741580825,
+      "required_pages_per_second_upper_estimate": 0.0022297759388543925
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28261.31413374527,
-      "required_pages_per_second_upper_estimate": 0.36250579307275776
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28680.412741580825,
+      "required_pages_per_second_upper_estimate": 0.3678815400210467
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06743975896085885,
-    "producer_p95_seconds": 0.21832736199939973,
-    "gpu_update_seconds": 0.9782984566519876,
-    "producer_batches_per_second": 14.828048252372712,
-    "gpu_batches_per_second": 1.0221829475457636,
-    "producer_headroom": 14.506256720457424,
+    "producer_batch_seconds": 0.08563265358597505,
+    "producer_p95_seconds": 0.23540296100327396,
+    "gpu_update_seconds": 0.9640028631776266,
+    "producer_batches_per_second": 11.67778829831545,
+    "gpu_batches_per_second": 1.0373413173314825,
+    "producer_headroom": 11.257421355158279,
     "sustainable": true,
-    "buffer_seconds": 60.65450431242323,
+    "buffer_seconds": 60.73218038019048,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 368866.18582722044,
-    "upload_bytes_per_second": 1917354.935290769,
+    "generation_bytes_per_second": 385520.0026931314,
+    "upload_bytes_per_second": 1902951.8234044153,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.15009671318492,
+    "estimated_full_shards_per_hour": 82.72361813159424,
     "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": 0.9640832400000363,
-      "bytes": 4456954,
-      "pages": 18,
-      "shards": 2
-    },
     {
       "seconds": 1.0857610709990695,
       "bytes": 6125796,
@@ -562,6 +556,12 @@
       "bytes": 7799504,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 0.9238773860015499,
+      "bytes": 1050494,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -187,3 +187,4 @@
 {"step": 663700, "loss": 0.5779117345809937, "lr": 0.0001, "grad_norm": 0.2095624644082802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97148526991, "tokens_per_second": 154575.90331443175, "data_wait_s": 1.5607998648192734e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 663800, "loss": 0.5807338356971741, "lr": 0.0001, "grad_norm": 0.1891549641286253, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97163369344, "tokens_per_second": 154110.84281717698, "data_wait_s": 1.8357000953983516e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 663900, "loss": 0.5550321340560913, "lr": 0.0001, "grad_norm": 0.2184161685445405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97178131334, "tokens_per_second": 154107.05724833082, "data_wait_s": 1.5232999430736527e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 664000, "loss": 0.5500065088272095, "lr": 0.0001, "grad_norm": 0.1818294185679898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97192694370, "tokens_per_second": 151407.37875300326, "data_wait_s": 9.307001164415851e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:149cbfc58150844a0651a7722e931f865c1abfadac48137012a4e39f4de4602d
-size 402816
+oid sha256:ed78497bd50637ee2f139a4ac8a37b86ca217817e363cc87fdd926f51a5f778a
+size 404934