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": 75468.78837771672,
-      "required_pages_per_second_upper_estimate": 0.0065414433026850525
+      "required_token_positions_per_second": 75466.30878295824,
+      "required_pages_per_second_upper_estimate": 0.006541228377695863
     },
     {
       "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": 9433.59854721459,
-      "required_pages_per_second_upper_estimate": 0.0037632035053512805
+      "required_token_positions_per_second": 9433.28859786978,
+      "required_pages_per_second_upper_estimate": 0.00376307986192348
     },
     {
       "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": 18867.19709442918,
-      "required_pages_per_second_upper_estimate": 0.00436459399646829
+      "required_token_positions_per_second": 18866.57719573956,
+      "required_pages_per_second_upper_estimate": 0.00436445059381629
     },
     {
       "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": 28300.79564164377,
-      "required_pages_per_second_upper_estimate": 0.0017233699717767292
+      "required_token_positions_per_second": 28299.86579360934,
+      "required_pages_per_second_upper_estimate": 0.0017233133489099682
     },
     {
       "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": 28300.79564164377,
-      "required_pages_per_second_upper_estimate": 0.00217867359011016
+      "required_token_positions_per_second": 28299.86579360934,
+      "required_pages_per_second_upper_estimate": 0.0021786020078344856
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28300.79564164377,
-      "required_pages_per_second_upper_estimate": 0.3514011651991478
+      "required_token_positions_per_second": 28299.86579360934,
+      "required_pages_per_second_upper_estimate": 0.3513896195985614
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.47765963539052336,
-    "producer_p95_seconds": 0.27693211199948564,
-    "gpu_update_seconds": 0.9769336646958716,
-    "producer_batches_per_second": 2.0935409356547856,
-    "gpu_batches_per_second": 1.0236109534738054,
-    "producer_headroom": 2.0452506184600536,
+    "producer_batch_seconds": 0.0896996522814959,
+    "producer_p95_seconds": 0.288132810001116,
+    "gpu_update_seconds": 0.9769657637826487,
+    "producer_batches_per_second": 11.148315233840538,
+    "gpu_batches_per_second": 1.0235773218174673,
+    "producer_headroom": 10.891522307318759,
     "sustainable": true,
-    "buffer_seconds": 60.56988721114404,
+    "buffer_seconds": 60.57187735452422,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 293528.80827824643,
-    "upload_bytes_per_second": 5227806.295094436,
+    "generation_bytes_per_second": 293179.9408926977,
+    "upload_bytes_per_second": 5255390.5287232315,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.98444925556702,
+    "estimated_full_shards_per_hour": 62.90959043584536,
     "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": 1.1885242320004181,
-      "bytes": 7012163,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.3779209780004749,
       "bytes": 11061746,
@@ -562,6 +556,12 @@
       "bytes": 7965202,
       "pages": 25,
       "shards": 3
+    },
+    {
+      "seconds": 1.4521252480008116,
+      "bytes": 10591031,
+      "pages": 45,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -204,3 +204,4 @@
 {"step": 215400, "loss": 0.5157265942543745, "lr": 0.0003, "grad_norm": 0.1502006032714767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31527713895, "tokens_per_second": 148299.84229734525, "data_wait_s": 7.127997378120199e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 215500, "loss": 0.6108162812888622, "lr": 0.0003, "grad_norm": 0.16144090286693072, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31542436363, "tokens_per_second": 133311.82707342232, "data_wait_s": 1.756699930410832e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 215600, "loss": 0.6187153682112694, "lr": 0.0003, "grad_norm": 0.162110905284232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31557255926, "tokens_per_second": 150366.4233226348, "data_wait_s": 1.19420001283288e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 215700, "loss": 0.614033866673708, "lr": 0.0003, "grad_norm": 0.1589129658866797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31571868086, "tokens_per_second": 147939.52947717562, "data_wait_s": 1.1487001756904647e-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:ba1165f12b6de3da23388696de6003976effae633e841ed5e709a438a2813dfb
-size 439605
+oid sha256:109233a05e754240963c179537ca34b1491fc74a8a1ae17e50d2b46e40edd9e0
+size 441723