Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77417.56951236352,
-      "required_pages_per_second_upper_estimate": 0.005355927092375245
+      "required_token_positions_per_second": 76981.04517608859,
+      "required_pages_per_second_upper_estimate": 0.0053257273259157386
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9677.19618904544,
-      "required_pages_per_second_upper_estimate": 0.004051639961517293
+      "required_token_positions_per_second": 9622.630647011074,
+      "required_pages_per_second_upper_estimate": 0.004028794508525594
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19354.39237809088,
-      "required_pages_per_second_upper_estimate": 0.0040310095227057575
+      "required_token_positions_per_second": 19245.261294022148,
+      "required_pages_per_second_upper_estimate": 0.004008280395874464
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29031.58856713632,
-      "required_pages_per_second_upper_estimate": 0.0017403994059307936
+      "required_token_positions_per_second": 28867.891941033224,
+      "required_pages_per_second_upper_estimate": 0.0017305860431461883
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29031.58856713632,
-      "required_pages_per_second_upper_estimate": 0.0022570783146321203
+      "required_token_positions_per_second": 28867.891941033224,
+      "required_pages_per_second_upper_estimate": 0.002244351621977968
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29031.58856713632,
-      "required_pages_per_second_upper_estimate": 0.3723860464480487
+      "required_token_positions_per_second": 28867.891941033224,
+      "required_pages_per_second_upper_estimate": 0.3702863218921412
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.19084989946096442,
-    "producer_p95_seconds": 0.14990303800004767,
-    "gpu_update_seconds": 0.95234196144876,
-    "producer_batches_per_second": 5.239719815542976,
-    "gpu_batches_per_second": 1.050042989262743,
-    "producer_headroom": 4.990005046576133,
+    "producer_batch_seconds": 0.04395073877342881,
+    "producer_p95_seconds": 0.14269972400006736,
+    "gpu_update_seconds": 0.9577422576083828,
+    "producer_batches_per_second": 22.752746095011435,
+    "gpu_batches_per_second": 1.0441222490246391,
+    "producer_headroom": 21.791266411826566,
     "sustainable": true,
-    "buffer_seconds": 60.94988553272064,
+    "buffer_seconds": 60.33776222932812,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430056.9315278553,
-    "upload_bytes_per_second": 3462370.632841698,
+    "generation_bytes_per_second": 364370.7712736338,
+    "upload_bytes_per_second": 3891508.561836774,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.28020629288429,
+    "estimated_full_shards_per_hour": 78.18548539788019,
     "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": [
@@ -280,6 +280,12 @@
       "bytes": 1085324,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 1.6373932590004188,
+      "bytes": 17208473,
+      "pages": 36,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -18,3 +18,4 @@
 {"step": 844300, "loss": 0.5937459468841553, "lr": 3e-05, "grad_norm": 0.31839938163663495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123581846732, "tokens_per_second": 151976.31945156935, "data_wait_s": 8.49900015964522e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 844400, "loss": 0.5504328608512878, "lr": 3e-05, "grad_norm": 0.27842239346038167, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123596652416, "tokens_per_second": 154180.72814609038, "data_wait_s": 8.850999620335642e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 844500, "loss": 0.5093451142311096, "lr": 3e-05, "grad_norm": 0.29406922530866353, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123611407048, "tokens_per_second": 151968.61987264256, "data_wait_s": 1.2919999790028669e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 844600, "loss": 0.5306952595710754, "lr": 3e-05, "grad_norm": 0.3029432781467202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123625929123, "tokens_per_second": 150448.66103590364, "data_wait_s": 1.7395000213582534e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6b4ffacaed6697bf783af7e2c870e4aa2f5f703cab86f9764aac44256fb7db91
-size 42525
+oid sha256:7872e04a43a46af37025921f83d9fe0aefe2bbfee64c7965c097062e3c3b8583
+size 44643