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": 75957.33837170151,
-      "required_pages_per_second_upper_estimate": 0.005254904913861197
+      "required_token_positions_per_second": 76684.96786803045,
+      "required_pages_per_second_upper_estimate": 0.005305244010750272
     },
     {
       "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": 9494.667296462689,
-      "required_pages_per_second_upper_estimate": 0.003975218925829608
+      "required_token_positions_per_second": 9585.620983503806,
+      "required_pages_per_second_upper_estimate": 0.004013299335264758
     },
     {
       "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": 18989.334592925377,
-      "required_pages_per_second_upper_estimate": 0.0039549776132769806
+      "required_token_positions_per_second": 19171.241967007612,
+      "required_pages_per_second_upper_estimate": 0.003992864122078257
     },
     {
       "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": 28484.00188938807,
-      "required_pages_per_second_upper_estimate": 0.0017075724207162302
+      "required_token_positions_per_second": 28756.86295051142,
+      "required_pages_per_second_upper_estimate": 0.0017239300246958585
     },
     {
       "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": 28484.00188938807,
-      "required_pages_per_second_upper_estimate": 0.002214505859016443
+      "required_token_positions_per_second": 28756.86295051142,
+      "required_pages_per_second_upper_estimate": 0.002235719606329815
     },
     {
       "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": 28484.00188938807,
-      "required_pages_per_second_upper_estimate": 0.3653621924986605
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28756.86295051142,
+      "required_pages_per_second_upper_estimate": 0.36886216121536947
     }
   ],
   "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": 0.18183362752330368,
-    "producer_p95_seconds": 0.3297312499998952,
-    "gpu_update_seconds": 0.9706501251953811,
-    "producer_batches_per_second": 5.499532807108744,
-    "gpu_batches_per_second": 1.0302373368557607,
-    "producer_headroom": 5.338122207736208,
+    "producer_batch_seconds": 0.0900163293275682,
+    "producer_p95_seconds": 0.3198437329992885,
+    "gpu_update_seconds": 0.9614400585898505,
+    "producer_batches_per_second": 11.10909551044915,
+    "gpu_batches_per_second": 1.0401064435225484,
+    "producer_headroom": 10.680729438446475,
     "sustainable": true,
-    "buffer_seconds": 60.18030776211363,
+    "buffer_seconds": 60.57072369116058,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 272930.16616160807,
-    "upload_bytes_per_second": 4586202.791331223,
+    "generation_bytes_per_second": 272940.89750342973,
+    "upload_bytes_per_second": 4650465.0101744365,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.564460169183555,
+    "estimated_full_shards_per_hour": 58.56676286532563,
     "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": 1.220427556996583,
-      "bytes": 2558602,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.1153945629994269,
       "bytes": 6138686,
@@ -562,6 +556,12 @@
       "bytes": 2763678,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.6161050989976502,
+      "bytes": 10229601,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -407,3 +407,4 @@
 {"step": 760700, "loss": 0.45216894149780273, "lr": 5e-05, "grad_norm": 0.24528988115250652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111344364855, "tokens_per_second": 152424.21800224096, "data_wait_s": 1.9842998881358653e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 760800, "loss": 0.535881519317627, "lr": 5e-05, "grad_norm": 0.24302539573863297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111358936214, "tokens_per_second": 150784.03338170476, "data_wait_s": 1.6279002011287957e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 760900, "loss": 0.5553132891654968, "lr": 5e-05, "grad_norm": 0.3694032271862332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111373484123, "tokens_per_second": 150724.3185302542, "data_wait_s": 1.548200089018792e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 761000, "loss": 0.5758531093597412, "lr": 5e-05, "grad_norm": 0.2517351415603714, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111387964808, "tokens_per_second": 149993.37426602026, "data_wait_s": 1.098900247598067e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1ef80e69912f72b44482c53980ef556fbcb0d5fa6bab2b11be6c42f05fe9b922
-size 872691
+oid sha256:fc2fb22d7a75a0939b69fe824de9442d68bcd17d8a5e4b4bb80a679ba70b47ec
+size 874809