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": 76966.941084021,
-      "required_pages_per_second_upper_estimate": 0.0053247515721005
+      "required_token_positions_per_second": 77479.51542505863,
+      "required_pages_per_second_upper_estimate": 0.0053602126543498135
     },
     {
       "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": 9620.867635502625,
-      "required_pages_per_second_upper_estimate": 0.004028056372422873
+      "required_token_positions_per_second": 9684.93942813233,
+      "required_pages_per_second_upper_estimate": 0.004054881894025754
     },
     {
       "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": 19241.73527100525,
-      "required_pages_per_second_upper_estimate": 0.004007546018267524
+      "required_token_positions_per_second": 19369.87885626466,
+      "required_pages_per_second_upper_estimate": 0.004034234947703462
     },
     {
       "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": 28862.602906507876,
-      "required_pages_per_second_upper_estimate": 0.001730268973602282
+      "required_token_positions_per_second": 29054.818284396988,
+      "required_pages_per_second_upper_estimate": 0.0017417919920108462
     },
     {
       "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": 28862.602906507876,
-      "required_pages_per_second_upper_estimate": 0.002243940422807627
+      "required_token_positions_per_second": 29054.818284396988,
+      "required_pages_per_second_upper_estimate": 0.0022588843229724093
     },
     {
       "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": 28862.602906507876,
-      "required_pages_per_second_upper_estimate": 0.3702184798361729
+      "required_token_positions_per_second": 29054.818284396988,
+      "required_pages_per_second_upper_estimate": 0.3726840123189414
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07467568585936846,
-    "producer_p95_seconds": 0.17311898499974632,
-    "gpu_update_seconds": 0.9579177626341521,
-    "producer_batches_per_second": 13.391239578076734,
-    "gpu_batches_per_second": 1.043930950032837,
-    "producer_headroom": 12.827706255529172,
+    "producer_batch_seconds": 0.09322337186712559,
+    "producer_p95_seconds": 0.1731212519998735,
+    "gpu_update_seconds": 0.9515805512660019,
+    "producer_batches_per_second": 10.726923731372146,
+    "gpu_batches_per_second": 1.0508831844761644,
+    "producer_headroom": 10.207531997687465,
     "sustainable": true,
-    "buffer_seconds": 60.34881904595158,
+    "buffer_seconds": 60.901155281024124,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 328291.84011425276,
-    "upload_bytes_per_second": 4800932.689696059,
+    "generation_bytes_per_second": 327638.75158359826,
+    "upload_bytes_per_second": 4818867.077246846,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.44378664560973,
+    "estimated_full_shards_per_hour": 70.30364905005418,
     "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": [
@@ -526,6 +526,12 @@
       "bytes": 1138477,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.3184527379999054,
+      "bytes": 7724848,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -68,3 +68,4 @@
 {"step": 403900, "loss": 0.563493013381958, "lr": 0.0003, "grad_norm": 0.157540287331852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59120453389, "tokens_per_second": 122755.2944953602, "data_wait_s": 9.309999768447597e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 404000, "loss": 0.576484203338623, "lr": 0.0003, "grad_norm": 0.17450364148191946, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59135043101, "tokens_per_second": 151872.57447063093, "data_wait_s": 1.0945999747491442e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 404100, "loss": 0.533147394657135, "lr": 0.0003, "grad_norm": 0.15824127040319796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59149658570, "tokens_per_second": 152590.37667120722, "data_wait_s": 1.0043000656878576e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 404200, "loss": 0.5693563222885132, "lr": 0.0003, "grad_norm": 0.15882381498968365, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59164350744, "tokens_per_second": 152735.23199207423, "data_wait_s": 1.4741000086360145e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:35839408821d6610ef7e0f727dad644c3e641a2457095ccf53632f7f2fcf00cc
-size 149208
+oid sha256:c4879d6314b5a4b58a00dfab159fa7bbf99c78ad7f90af876384acb9593561d1
+size 151326