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": 74689.00920818429,
-      "required_pages_per_second_upper_estimate": 0.006473854020602218
+      "required_token_positions_per_second": 75229.4140338592,
+      "required_pages_per_second_upper_estimate": 0.006520694941248211
     },
     {
       "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": 9336.126151023036,
-      "required_pages_per_second_upper_estimate": 0.0037243203091682766
+      "required_token_positions_per_second": 9403.6767542324,
+      "required_pages_per_second_upper_estimate": 0.0037512672547600125
     },
     {
       "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": 18672.252302046072,
-      "required_pages_per_second_upper_estimate": 0.004319496949661624
+      "required_token_positions_per_second": 18807.3535084648,
+      "required_pages_per_second_upper_estimate": 0.004350750236066572
     },
     {
       "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": 28008.37845306911,
-      "required_pages_per_second_upper_estimate": 0.0017055632991869521
+      "required_token_positions_per_second": 28211.0302626972,
+      "required_pages_per_second_upper_estimate": 0.0017179037311614287
     },
     {
       "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": 28008.37845306911,
-      "required_pages_per_second_upper_estimate": 0.0021561625054710965
+      "required_token_positions_per_second": 28211.0302626972,
+      "required_pages_per_second_upper_estimate": 0.00217176320275238
     },
     {
       "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": 28008.37845306911,
-      "required_pages_per_second_upper_estimate": 0.3477703223744255
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28211.0302626972,
+      "required_pages_per_second_upper_estimate": 0.350286579618029
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08218930775777267,
-    "producer_p95_seconds": 0.31462730600105715,
-    "gpu_update_seconds": 0.9871331911030494,
-    "producer_batches_per_second": 12.167032759871732,
-    "gpu_batches_per_second": 1.0130345215953815,
-    "producer_headroom": 12.010481874507526,
+    "producer_batch_seconds": 0.24345651046877492,
+    "producer_p95_seconds": 0.31245939200016437,
+    "gpu_update_seconds": 0.9800421942249418,
+    "producer_batches_per_second": 4.10750978922068,
+    "gpu_batches_per_second": 1.0203642311450087,
+    "producer_headroom": 4.025532906628263,
     "sustainable": true,
-    "buffer_seconds": 60.21512465728601,
+    "buffer_seconds": 60.76261604194639,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 517914.29029247066,
-    "upload_bytes_per_second": 5254492.497796874,
+    "generation_bytes_per_second": 493625.461825298,
+    "upload_bytes_per_second": 5194047.996868255,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.13235026913252,
+    "estimated_full_shards_per_hour": 105.92053309506612,
     "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": 1.0304640750000544,
-      "bytes": 6585841,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.9793735439998272,
       "bytes": 7348514,
@@ -562,6 +556,12 @@
       "bytes": 6164642,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 0.908005076002155,
+      "bytes": 1082411,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -255,3 +255,4 @@
 {"step": 340500, "loss": 0.6149112656712532, "lr": 0.0003, "grad_norm": 0.1481979501688307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49840024880, "tokens_per_second": 150326.8567130313, "data_wait_s": 1.1689000530168414e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 340600, "loss": 0.5959947258234024, "lr": 0.0003, "grad_norm": 0.14344431628262702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49854778812, "tokens_per_second": 150351.13342953153, "data_wait_s": 1.4061002730159089e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 340700, "loss": 0.6126907952129841, "lr": 0.0003, "grad_norm": 0.16114369116465935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49869487722, "tokens_per_second": 150051.98153942704, "data_wait_s": 1.6572001186432317e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 340800, "loss": 0.6215404812246561, "lr": 0.0003, "grad_norm": 0.18114448725654078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49884120556, "tokens_per_second": 149051.0557850749, "data_wait_s": 1.3945002137916163e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ed5d51203a3e729b36b53b344ad5a651151929916fff09dfe9eb08bf00ca1f07
-size 548406
+oid sha256:38f7a5e4337d9e873e85d2bc87a2ab9d3e49b89c5d920c8a2d7212f1acf4b0a6
+size 550524