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": 77296.81051698115,
-      "required_pages_per_second_upper_estimate": 0.005347572704875219
+      "required_token_positions_per_second": 76370.36274281703,
+      "required_pages_per_second_upper_estimate": 0.005283478898203545
     },
     {
       "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": 9662.101314622643,
-      "required_pages_per_second_upper_estimate": 0.004045320052813291
+      "required_token_positions_per_second": 9546.295342852129,
+      "required_pages_per_second_upper_estimate": 0.003996834510736661
     },
     {
       "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": 19324.202629245287,
-      "required_pages_per_second_upper_estimate": 0.004024721794178437
+      "required_token_positions_per_second": 19092.590685704257,
+      "required_pages_per_second_upper_estimate": 0.003976483134356538
     },
     {
       "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": 28986.303943867933,
-      "required_pages_per_second_upper_estimate": 0.0017376846619114698
+      "required_token_positions_per_second": 28638.88602855639,
+      "required_pages_per_second_upper_estimate": 0.0017168574883649839
     },
     {
       "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": 28986.303943867933,
-      "required_pages_per_second_upper_estimate": 0.0022535576343590104
+      "required_token_positions_per_second": 28638.88602855639,
+      "required_pages_per_second_upper_estimate": 0.0022265474195734495
     },
     {
       "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": 28986.303943867933,
-      "required_pages_per_second_upper_estimate": 0.37180518392360196
+      "required_token_positions_per_second": 28638.88602855639,
+      "required_pages_per_second_upper_estimate": 0.36734887993427984
     }
   ],
   "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": 0.2255064647265641,
-    "producer_p95_seconds": 0.23226359399996,
-    "gpu_update_seconds": 0.9538297829740707,
-    "producer_batches_per_second": 4.434462671447319,
-    "gpu_batches_per_second": 1.0484050905623528,
-    "producer_headroom": 4.229722567513214,
+    "producer_batch_seconds": 0.11466585946089936,
+    "producer_p95_seconds": 0.2418848990000697,
+    "gpu_update_seconds": 0.9654006783794469,
+    "producer_batches_per_second": 8.720991624721536,
+    "gpu_batches_per_second": 1.0358393384171147,
+    "producer_headroom": 8.419251230647646,
     "sustainable": true,
-    "buffer_seconds": 60.09127632736646,
+    "buffer_seconds": 60.820242737905154,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 494611.81918314396,
-    "upload_bytes_per_second": 2159870.8971751374,
+    "generation_bytes_per_second": 474838.37521415454,
+    "upload_bytes_per_second": 1872218.810960124,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.13218242283573,
+    "estimated_full_shards_per_hour": 101.88926164930798,
     "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": 1.255959656999039,
-      "bytes": 8183161,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.190589429001193,
       "bytes": 6346216,
@@ -562,6 +556,12 @@
       "bytes": 5676510,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 33.502104078999764,
+      "bytes": 11160724,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -159,3 +159,4 @@
 {"step": 660900, "loss": 0.5548530220985413, "lr": 0.0001, "grad_norm": 0.18112856739384112, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96738195214, "tokens_per_second": 151036.16568345609, "data_wait_s": 8.747998435865156e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 661000, "loss": 0.5415875315666199, "lr": 0.0001, "grad_norm": 0.19328856622939006, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96752770905, "tokens_per_second": 152089.57396724008, "data_wait_s": 8.262999472208321e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 661100, "loss": 0.5500502586364746, "lr": 0.0001, "grad_norm": 0.19395241358893767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96767415798, "tokens_per_second": 152284.56853286334, "data_wait_s": 1.1494001228129491e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 661200, "loss": 0.5655838847160339, "lr": 0.0001, "grad_norm": 0.18529066629640334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96782162489, "tokens_per_second": 154032.61333851627, "data_wait_s": 8.934999641496688e-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 sha256:4e3028b3646c9fb925a3d3a0015d6a44651d4231b56b8c71cd887813732de0f6
-size 343512
+oid sha256:1f876269aa8bee2dd6998550391c667204ec249560ea14b5d111073dac237b20
+size 345630