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": 75284.34489320319,
-      "required_pages_per_second_upper_estimate": 0.0065254562089151575
+      "required_token_positions_per_second": 75917.11416502793,
+      "required_pages_per_second_upper_estimate": 0.006580303045657869
     },
     {
       "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": 9410.543111650399,
-      "required_pages_per_second_upper_estimate": 0.0037540063473952444
+      "required_token_positions_per_second": 9489.639270628491,
+      "required_pages_per_second_upper_estimate": 0.0037855589878045682
     },
     {
       "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": 18821.086223300797,
-      "required_pages_per_second_upper_estimate": 0.004353927057956354
+      "required_token_positions_per_second": 18979.278541256983,
+      "required_pages_per_second_upper_estimate": 0.004390522066625809
     },
     {
       "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": 28231.629334951194,
-      "required_pages_per_second_upper_estimate": 0.0017191581065867172
+      "required_token_positions_per_second": 28468.917811885472,
+      "required_pages_per_second_upper_estimate": 0.0017336077298755915
     },
     {
       "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": 28231.629334951194,
-      "required_pages_per_second_upper_estimate": 0.0021733489763563745
+      "required_token_positions_per_second": 28468.917811885472,
+      "required_pages_per_second_upper_estimate": 0.0021916161001673197
     },
     {
       "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": 28231.629334951194,
-      "required_pages_per_second_upper_estimate": 0.35054235115476357
+      "required_token_positions_per_second": 28468.917811885472,
+      "required_pages_per_second_upper_estimate": 0.3534886798848414
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.44732809546099617,
-    "producer_p95_seconds": 0.21524411300015345,
-    "gpu_update_seconds": 0.9793271111622079,
-    "producer_batches_per_second": 2.2354956242340314,
-    "gpu_batches_per_second": 1.0211092786078992,
-    "producer_headroom": 2.1892814716968707,
+    "producer_batch_seconds": 0.05740078666403292,
+    "producer_p95_seconds": 0.20062419999885606,
+    "gpu_update_seconds": 0.9711644180748329,
+    "producer_batches_per_second": 17.421364028562966,
+    "gpu_batches_per_second": 1.029691761135904,
+    "producer_headroom": 16.91900885886918,
     "sustainable": true,
-    "buffer_seconds": 60.718280892056896,
+    "buffer_seconds": 60.21219392063964,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320208.4713715313,
-    "upload_bytes_per_second": 5786423.735652191,
+    "generation_bytes_per_second": 319337.8969382163,
+    "upload_bytes_per_second": 5930099.599497368,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.70928388461546,
+    "estimated_full_shards_per_hour": 68.52247887716166,
     "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": [
@@ -328,6 +328,12 @@
       "bytes": 2564629,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.0954296749987407,
+      "bytes": 10402230,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -77,3 +77,4 @@
 {"step": 122700, "loss": 0.6094257552176714, "lr": 0.0003, "grad_norm": 0.13699610787656766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17955682394, "tokens_per_second": 148236.24361761165, "data_wait_s": 1.1531999916769564e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 122800, "loss": 0.5962003283202648, "lr": 0.0003, "grad_norm": 0.1375012183929598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17970401412, "tokens_per_second": 149914.04022148447, "data_wait_s": 1.2873999367002398e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 122900, "loss": 0.6043729484081268, "lr": 0.0003, "grad_norm": 0.13227822076269233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17985221717, "tokens_per_second": 151421.41966565498, "data_wait_s": 1.3359998774831183e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 123000, "loss": 0.6179828643798828, "lr": 0.0003, "grad_norm": 0.13953831270701456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17999883207, "tokens_per_second": 149125.69639491214, "data_wait_s": 1.533800059405621e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:61752383b32629c701d13972174074503ef482941a192ffbb9b30631333af5ef
-size 168270
+oid sha256:9b4954dbb23dad2e29a60a5f1d89a75bd7fd09afc34b8e257bd016198ed0a167
+size 170388