Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 73481.42210036334,
-      "required_pages_per_second_upper_estimate": 0.0063691834306978425
+      "required_token_positions_per_second": 74455.14508968488,
+      "required_pages_per_second_upper_estimate": 0.006453583271533879
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9185.177762545418,
-      "required_pages_per_second_upper_estimate": 0.003664104740125027
+      "required_token_positions_per_second": 9306.89313621061,
+      "required_pages_per_second_upper_estimate": 0.0037126588224870795
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18370.355525090836,
-      "required_pages_per_second_upper_estimate": 0.004249658443514836
+      "required_token_positions_per_second": 18613.78627242122,
+      "required_pages_per_second_upper_estimate": 0.004305971862674892
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27555.53328763625,
-      "required_pages_per_second_upper_estimate": 0.0016779874045070552
+      "required_token_positions_per_second": 27920.679408631833,
+      "required_pages_per_second_upper_estimate": 0.0017002228875020481
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27555.53328763625,
-      "required_pages_per_second_upper_estimate": 0.0021213012310804264
+      "required_token_positions_per_second": 27920.679408631833,
+      "required_pages_per_second_upper_estimate": 0.0021494111902638224
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27555.53328763625,
-      "required_pages_per_second_upper_estimate": 0.34214750099502406
+      "required_token_positions_per_second": 27920.679408631833,
+      "required_pages_per_second_upper_estimate": 0.34668139375233537
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 60,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.41790056917190554,
-    "producer_p95_seconds": 0.31119969800056424,
-    "gpu_update_seconds": 1.003355649531386,
-    "producer_batches_per_second": 2.392913706678023,
-    "gpu_batches_per_second": 0.9966555731928621,
-    "producer_headroom": 2.4009434864364843,
+    "producer_batch_seconds": 0.2559345569219573,
+    "producer_p95_seconds": 0.2806939820002299,
+    "gpu_update_seconds": 0.9902337831884015,
+    "producer_batches_per_second": 3.9072488374632903,
+    "gpu_batches_per_second": 1.009862536481186,
+    "producer_headroom": 3.8690897981797576,
     "sustainable": true,
-    "buffer_seconds": 60.20133897188316,
+    "buffer_seconds": 60.40426077449249,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340717.94559244555,
-    "upload_bytes_per_second": 5215582.723506866,
+    "generation_bytes_per_second": 329478.1069865731,
+    "upload_bytes_per_second": 5356668.188638754,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.11013961629891,
+    "estimated_full_shards_per_hour": 70.69833190153021,
     "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": 0.9941408290005711,
-      "bytes": 1118169,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.4089016780008023,
       "bytes": 7539453,
@@ -562,6 +556,12 @@
       "bytes": 2531508,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.6972403340005258,
+      "bytes": 16066476,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -143,3 +143,4 @@
 {"step": 209300, "loss": 0.61921376734972, "lr": 0.0003, "grad_norm": 0.14681122395405305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30633559690, "tokens_per_second": 147517.66940943303, "data_wait_s": 1.7534999642521143e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 209400, "loss": 0.6080213040113449, "lr": 0.0003, "grad_norm": 0.14197261232353933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30648187358, "tokens_per_second": 147741.7675434638, "data_wait_s": 1.1895001080119982e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 209500, "loss": 0.6023000460118055, "lr": 0.0003, "grad_norm": 0.14871573819746792, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30662903361, "tokens_per_second": 149183.30527687844, "data_wait_s": 1.7944001228897832e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 209600, "loss": 0.616977671161294, "lr": 0.0003, "grad_norm": 0.1456166156245679, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30677633997, "tokens_per_second": 148708.33548059704, "data_wait_s": 1.6396999853895977e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0e9a751dfad2586abd86dccfed2794c2812395120912862d7794ae6ec537986c
-size 308841
+oid sha256:8b1541cabc317d7602c673db7b01d06ffd310cef928814203fe5471f8c791f4b
+size 310959