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": 75734.90116234886,
-      "required_pages_per_second_upper_estimate": 0.006564509284400281
+      "required_token_positions_per_second": 75516.61507881177,
+      "required_pages_per_second_upper_estimate": 0.006545588799920306
     },
     {
       "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": 9466.862645293608,
-      "required_pages_per_second_upper_estimate": 0.0037764730514175874
+      "required_token_positions_per_second": 9439.57688485147,
+      "required_pages_per_second_upper_estimate": 0.0037655883536187452
     },
     {
       "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": 18933.725290587216,
-      "required_pages_per_second_upper_estimate": 0.004379984123793188
+      "required_token_positions_per_second": 18879.15376970294,
+      "required_pages_per_second_upper_estimate": 0.0043673599628625765
     },
     {
       "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": 28400.587935880823,
-      "required_pages_per_second_upper_estimate": 0.0017294467989260615
+      "required_token_positions_per_second": 28318.730654554412,
+      "required_pages_per_second_upper_estimate": 0.001724462119965247
     },
     {
       "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": 28400.587935880823,
-      "required_pages_per_second_upper_estimate": 0.002186355877163279
+      "required_token_positions_per_second": 28318.730654554412,
+      "required_pages_per_second_upper_estimate": 0.0021800542770513145
     },
     {
       "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": 28400.587935880823,
-      "required_pages_per_second_upper_estimate": 0.35264025151024775
+      "required_token_positions_per_second": 28318.730654554412,
+      "required_pages_per_second_upper_estimate": 0.35162385803487106
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.09514931401542981,
-    "producer_p95_seconds": 0.37020335299894214,
-    "gpu_update_seconds": 0.9735009733749203,
-    "producer_batches_per_second": 10.509797262835082,
-    "gpu_batches_per_second": 1.02722033911606,
-    "producer_headroom": 10.231297865343025,
+    "producer_batch_seconds": 0.14390511967206976,
+    "producer_p95_seconds": 0.388960545998998,
+    "gpu_update_seconds": 0.9763149463605445,
+    "producer_batches_per_second": 6.949023094374924,
+    "gpu_batches_per_second": 1.0242596446236405,
+    "producer_headroom": 6.784435109642839,
     "sustainable": true,
-    "buffer_seconds": 60.35706034924505,
+    "buffer_seconds": 60.531526674353756,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 313234.4895998089,
-    "upload_bytes_per_second": 5232234.624428951,
+    "generation_bytes_per_second": 314242.27369879343,
+    "upload_bytes_per_second": 5183179.792680808,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.21282974239062,
+    "estimated_full_shards_per_hour": 67.42907675001958,
     "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": 1.3798298289984814,
-      "bytes": 7738436,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.956485711998539,
       "bytes": 1216085,
@@ -562,6 +556,12 @@
       "bytes": 10099926,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.8219810020018485,
+      "bytes": 940002,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -371,3 +371,4 @@
 {"step": 274200, "loss": 0.6100579909980297, "lr": 0.0003, "grad_norm": 0.1579126367697439, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40134537871, "tokens_per_second": 147906.9049218179, "data_wait_s": 1.1185002222191542e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 274300, "loss": 0.5829500239342451, "lr": 0.0003, "grad_norm": 0.12775942397489143, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40148945799, "tokens_per_second": 147073.15038042015, "data_wait_s": 1.310199877480045e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 274400, "loss": 0.5805154629051685, "lr": 0.0003, "grad_norm": 0.14057968008857563, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40163496188, "tokens_per_second": 148212.90861629412, "data_wait_s": 1.2228993000462651e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 274500, "loss": 0.5555185135453939, "lr": 0.0003, "grad_norm": 0.14979814296883148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40178122704, "tokens_per_second": 148706.46688129418, "data_wait_s": 1.054699532687664e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3aa1c33d0c179b8180984b627d8f1db2b7c1d9e5b3c4d53d09a60e7f091bb104
-size 795660
+oid sha256:8e069bdbeccdfcc91123e6d6ef6e22467d600e52b6c25d399bdf8ebec6cbc011
+size 797778