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": 75374.19491799061,
-      "required_pages_per_second_upper_estimate": 0.0065332441813409255
+      "required_token_positions_per_second": 74546.33734906974,
+      "required_pages_per_second_upper_estimate": 0.006461487585452691
     },
     {
       "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": 9421.774364748826,
-      "required_pages_per_second_upper_estimate": 0.003758486662178405
+      "required_token_positions_per_second": 9318.292168633718,
+      "required_pages_per_second_upper_estimate": 0.003717206066951379
     },
     {
       "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": 18843.548729497652,
-      "required_pages_per_second_upper_estimate": 0.004359123363438389
+      "required_token_positions_per_second": 18636.584337267435,
+      "required_pages_per_second_upper_estimate": 0.004311245793744826
     },
     {
       "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": 28265.323094246476,
-      "required_pages_per_second_upper_estimate": 0.001721209879750307
+      "required_token_positions_per_second": 27954.876505901157,
+      "required_pages_per_second_upper_estimate": 0.0017023053113074467
     },
     {
       "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": 28265.323094246476,
-      "required_pages_per_second_upper_estimate": 0.0021759428152172203
+      "required_token_positions_per_second": 27954.876505901157,
+      "required_pages_per_second_upper_estimate": 0.002152043777475239
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28265.323094246476,
-      "required_pages_per_second_upper_estimate": 0.35096071487945263
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27954.876505901157,
+      "required_pages_per_second_upper_estimate": 0.34710600725009816
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.06056848967969586,
-    "producer_p95_seconds": 0.2017865879997771,
-    "gpu_update_seconds": 0.9781597014763247,
-    "producer_batches_per_second": 16.510235029605273,
-    "gpu_batches_per_second": 1.0223279475638916,
-    "producer_headroom": 16.149646567862654,
+    "producer_batch_seconds": 0.388744507273465,
+    "producer_p95_seconds": 0.21869220700045844,
+    "gpu_update_seconds": 0.9890224338556325,
+    "producer_batches_per_second": 2.5723836126038,
+    "gpu_batches_per_second": 1.0110994106590405,
+    "producer_headroom": 2.5441451013477545,
     "sustainable": true,
-    "buffer_seconds": 60.64590149153213,
+    "buffer_seconds": 60.33036846519358,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 310311.4499355704,
-    "upload_bytes_per_second": 3612803.896304032,
+    "generation_bytes_per_second": 295417.628423651,
+    "upload_bytes_per_second": 3628441.4914039075,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.58561347532591,
+    "estimated_full_shards_per_hour": 63.38974608928821,
     "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": [
@@ -544,6 +544,12 @@
       "bytes": 10729274,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.2665546050002376,
+      "bytes": 6293373,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -73,3 +73,4 @@
 {"step": 244400, "loss": 0.6831602342426777, "lr": 0.0003, "grad_norm": 0.19541405455019706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35772676862, "tokens_per_second": 138980.74329869053, "data_wait_s": 1.78360005520517e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 244500, "loss": 0.5546451276168227, "lr": 0.0003, "grad_norm": 0.15389539280029582, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35787392726, "tokens_per_second": 149817.3266908133, "data_wait_s": 1.715200050966814e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 244600, "loss": 0.6167989317327738, "lr": 0.0003, "grad_norm": 0.14981602793378007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35802099670, "tokens_per_second": 149243.78627961257, "data_wait_s": 1.713599976937985e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 244700, "loss": 0.6106148473918438, "lr": 0.0003, "grad_norm": 0.16182671390144418, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35816796630, "tokens_per_second": 148865.4245083256, "data_wait_s": 1.6652999875077512e-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 sha256:49cc25fc310223aa6d2a3587ecd122c6e89cf0d310134ff7b067d5d7823f0c94
-size 159798
+oid sha256:4273b0419514589ea13f446fc5d5ca2637c41570bbbb39f0ce5b3c6cf18a3e52
+size 161916