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": 72751.05984067393,
-      "required_pages_per_second_upper_estimate": 0.006305877481114188
+      "required_token_positions_per_second": 75188.63646231969,
+      "required_pages_per_second_upper_estimate": 0.006517160444696976
     },
     {
       "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": 9093.882480084241,
-      "required_pages_per_second_upper_estimate": 0.0036276856869651513
+      "required_token_positions_per_second": 9398.57955778996,
+      "required_pages_per_second_upper_estimate": 0.0037492339068892454
     },
     {
       "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": 18187.764960168483,
-      "required_pages_per_second_upper_estimate": 0.004207419329804238
+      "required_token_positions_per_second": 18797.15911557992,
+      "required_pages_per_second_upper_estimate": 0.004348391942687842
     },
     {
       "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": 27281.647440252724,
-      "required_pages_per_second_upper_estimate": 0.0016613091933693849
+      "required_token_positions_per_second": 28195.738673369884,
+      "required_pages_per_second_upper_estimate": 0.0017169725536001637
     },
     {
       "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": 27281.647440252724,
-      "required_pages_per_second_upper_estimate": 0.0021002167403842923
+      "required_token_positions_per_second": 28195.738673369884,
+      "required_pages_per_second_upper_estimate": 0.002170586013876135
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 27281.647440252724,
-      "required_pages_per_second_upper_estimate": 0.3387467554074863
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28195.738673369884,
+      "required_pages_per_second_upper_estimate": 0.35009670925624103
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.13885986795344252,
-    "producer_p95_seconds": 0.4091569150041323,
-    "gpu_update_seconds": 1.0134285350820398,
-    "producer_batches_per_second": 7.201504759714188,
-    "gpu_batches_per_second": 0.9867494010508074,
-    "producer_headroom": 7.298210419023487,
+    "producer_batch_seconds": 0.20874050276592016,
+    "producer_p95_seconds": 0.39440612599719316,
+    "gpu_update_seconds": 0.9805737072642396,
+    "producer_batches_per_second": 4.79063711521952,
+    "gpu_batches_per_second": 1.0198111499338065,
+    "producer_headroom": 4.697572796228467,
     "sustainable": true,
-    "buffer_seconds": 60.805712104922385,
+    "buffer_seconds": 60.79556985038286,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 423760.89114107436,
-    "upload_bytes_per_second": 5682246.595111373,
+    "generation_bytes_per_second": 429269.96778147056,
+    "upload_bytes_per_second": 5639328.057286272,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.92922259020017,
+    "estimated_full_shards_per_hour": 92.11134219248855,
     "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": 1.035560734999308,
-      "bytes": 6144689,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3160131810000166,
       "bytes": 6821427,
@@ -562,6 +556,12 @@
       "bytes": 10625968,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 0.9398709419983788,
+      "bytes": 2450871,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -389,3 +389,4 @@
 {"step": 393900, "loss": 0.5727016478776932, "lr": 0.0003, "grad_norm": 0.15051099218322683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57655563623, "tokens_per_second": 146516.30754007335, "data_wait_s": 1.6882993804756552e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 394000, "loss": 0.5279850102961063, "lr": 0.0003, "grad_norm": 0.15053944217458426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57670155952, "tokens_per_second": 147895.81421272593, "data_wait_s": 1.1159005225636065e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 394100, "loss": 0.5695737525820732, "lr": 0.0003, "grad_norm": 0.14450641393979294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57684828750, "tokens_per_second": 149619.47986691375, "data_wait_s": 1.612699998077005e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 394200, "loss": 0.5794911608099937, "lr": 0.0003, "grad_norm": 0.14452503664774535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57699487255, "tokens_per_second": 149386.44426603787, "data_wait_s": 1.0415998985990882e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:10608d1b72ae0d86273c140502b740de6e8b61b31c71cea58926dc1beb2178f6
-size 833784
+oid sha256:4f9baf2ae53985ba765d445476aaa13f125285d311cfac6b7362a97326523333
+size 835902