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": 76336.46397358197,
-      "required_pages_per_second_upper_estimate": 0.005281133702691882
+      "required_token_positions_per_second": 76316.90624521802,
+      "required_pages_per_second_upper_estimate": 0.005279780653663482
     },
     {
       "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": 9542.057996697746,
-      "required_pages_per_second_upper_estimate": 0.003995060422387671
+      "required_token_positions_per_second": 9539.613280652253,
+      "required_pages_per_second_upper_estimate": 0.003994036870830899
     },
     {
       "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": 19084.11599339549,
-      "required_pages_per_second_upper_estimate": 0.003974718079441287
+      "required_token_positions_per_second": 19079.226561304506,
+      "required_pages_per_second_upper_estimate": 0.0039736997396797305
     },
     {
       "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": 28626.173990093237,
-      "required_pages_per_second_upper_estimate": 0.0017160954210692768
+      "required_token_positions_per_second": 28618.839841956757,
+      "required_pages_per_second_upper_estimate": 0.0017156557500870905
     },
     {
       "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": 28626.173990093237,
-      "required_pages_per_second_upper_estimate": 0.0022255591144972873
+      "required_token_positions_per_second": 28618.839841956757,
+      "required_pages_per_second_upper_estimate": 0.0022249889167391844
     },
     {
       "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": 28626.173990093237,
-      "required_pages_per_second_upper_estimate": 0.367185823553998
+      "required_token_positions_per_second": 28618.839841956757,
+      "required_pages_per_second_upper_estimate": 0.36709174897649793
     }
   ],
   "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": 0.10781931517185228,
-    "producer_p95_seconds": 0.24526225800036627,
-    "gpu_update_seconds": 0.9658293843099062,
-    "producer_batches_per_second": 9.274776030677886,
-    "gpu_batches_per_second": 1.035379556933349,
-    "producer_headroom": 8.957851223321898,
+    "producer_batch_seconds": 0.07009969477343247,
+    "producer_p95_seconds": 0.222215967000011,
+    "gpu_update_seconds": 0.9660768973404207,
+    "producer_batches_per_second": 14.265397349190689,
+    "gpu_batches_per_second": 1.0351142882652185,
+    "producer_headroom": 13.781470810434403,
     "sustainable": true,
-    "buffer_seconds": 60.84725121152409,
+    "buffer_seconds": 60.8628445324465,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 465570.4701467791,
-    "upload_bytes_per_second": 3488699.4030876844,
+    "generation_bytes_per_second": 466362.6698133397,
+    "upload_bytes_per_second": 3418995.7425089413,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.90058496763734,
+    "estimated_full_shards_per_hour": 100.07057257461685,
     "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": 1.2845682920014951,
-      "bytes": 11638802,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 1.3405057710006076,
       "bytes": 7718368,
@@ -562,6 +556,12 @@
       "bytes": 7075109,
       "pages": 43,
       "shards": 3
+    },
+    {
+      "seconds": 2.129564337999909,
+      "bytes": 5884538,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -158,3 +158,4 @@
 {"step": 858300, "loss": 0.5474139451980591, "lr": 3e-05, "grad_norm": 0.297106369941837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125631118251, "tokens_per_second": 150321.89091603263, "data_wait_s": 1.5887000699876808e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 858400, "loss": 0.5285065174102783, "lr": 3e-05, "grad_norm": 0.30141861417679533, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125645812232, "tokens_per_second": 153058.76738720742, "data_wait_s": 1.5947000065352768e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 858500, "loss": 0.49152708053588867, "lr": 3e-05, "grad_norm": 0.284207025313917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125660434830, "tokens_per_second": 151467.72231918594, "data_wait_s": 1.0134999683941714e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 858600, "loss": 0.5888546109199524, "lr": 3e-05, "grad_norm": 0.33249903731800506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125675099357, "tokens_per_second": 152094.5574051788, "data_wait_s": 1.5794999853824265e-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:c81f3218d4b0c45b0a75cf75523af860d5c0ff5b090915b63354fd9fdebb4ae2
-size 341394
+oid sha256:a5a1708fd554f57cbca41c7f3c0088585cbb003cf9e493193e98c704ea63f6b4
+size 343512