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": 74998.94668657484,
-      "required_pages_per_second_upper_estimate": 0.006500718615699754
+      "required_token_positions_per_second": 75306.70219319577,
+      "required_pages_per_second_upper_estimate": 0.006527394083014456
     },
     {
       "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": 9374.868335821855,
-      "required_pages_per_second_upper_estimate": 0.0037397751459318073
+      "required_token_positions_per_second": 9413.337774149471,
+      "required_pages_per_second_upper_estimate": 0.0037551211800500523
     },
     {
       "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": 18749.73667164371,
-      "required_pages_per_second_upper_estimate": 0.004337421594889708
+      "required_token_positions_per_second": 18826.675548298943,
+      "required_pages_per_second_upper_estimate": 0.004355220049925914
     },
     {
       "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": 28124.605007465565,
-      "required_pages_per_second_upper_estimate": 0.0017126408865560927
+      "required_token_positions_per_second": 28240.013322448413,
+      "required_pages_per_second_upper_estimate": 0.0017196686474379148
     },
     {
       "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": 28124.605007465565,
-      "required_pages_per_second_upper_estimate": 0.0021651099473642306
+      "required_token_positions_per_second": 28240.013322448413,
+      "required_pages_per_second_upper_estimate": 0.002173994398922273
     },
     {
       "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": 28124.605007465565,
-      "required_pages_per_second_upper_estimate": 0.34921346719477464
+      "required_token_positions_per_second": 28240.013322448413,
+      "required_pages_per_second_upper_estimate": 0.3506464522200779
     }
   ],
   "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": 0.10296336389077965,
-    "producer_p95_seconds": 0.32298557399917627,
-    "gpu_update_seconds": 0.9830538061836228,
-    "producer_batches_per_second": 9.712192397489742,
-    "gpu_batches_per_second": 1.0172383176890034,
-    "producer_headroom": 9.547607702739937,
+    "producer_batch_seconds": 0.08156905053931496,
+    "producer_p95_seconds": 0.3371588399968459,
+    "gpu_update_seconds": 0.9790363653271433,
+    "producer_batches_per_second": 12.259551795543045,
+    "gpu_batches_per_second": 1.021412518896427,
+    "producer_headroom": 12.002547030448316,
     "sustainable": true,
-    "buffer_seconds": 60.949335983384614,
+    "buffer_seconds": 60.700254650282886,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 422269.92770907824,
-    "upload_bytes_per_second": 4130836.3573462437,
+    "generation_bytes_per_second": 452879.79581134085,
+    "upload_bytes_per_second": 4122382.4611861496,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.60929654554616,
+    "estimated_full_shards_per_hour": 97.17746167903108,
     "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": 1.2729225670009328,
-      "bytes": 7447736,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.1847233869993943,
       "bytes": 10716114,
@@ -562,6 +556,12 @@
       "bytes": 2576063,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.1285306259960635,
+      "bytes": 5981374,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -243,3 +243,4 @@
 {"step": 109300, "loss": 0.6230886615812778, "lr": 0.0003, "grad_norm": 0.14055154783394633, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15994395784, "tokens_per_second": 148527.6360667483, "data_wait_s": 1.4556004316546023e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 109400, "loss": 0.5503679495304823, "lr": 0.0003, "grad_norm": 0.13035026920431542, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16009047196, "tokens_per_second": 148694.1929978418, "data_wait_s": 1.4409997675102204e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 109500, "loss": 0.5849812235683203, "lr": 0.0003, "grad_norm": 0.13039552184598213, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16023487806, "tokens_per_second": 146737.0379731941, "data_wait_s": 1.657399843679741e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 109600, "loss": 0.5687562748789787, "lr": 0.0003, "grad_norm": 0.1292022099549797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16038041293, "tokens_per_second": 137484.17929221195, "data_wait_s": 1.445900124963373e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2d2b0fcc4e499aca03266e27af341d95feb7497522e5df755952deaec483994f
-size 522207
+oid sha256:073e693f876641f711ff40618bb59a880c0aa024cba888d18fb71d726a4eb594
+size 524325