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": 76979.7982057961,
-      "required_pages_per_second_upper_estimate": 0.005325641057617534
+      "required_token_positions_per_second": 77170.93269939696,
+      "required_pages_per_second_upper_estimate": 0.005338864185377968
     },
     {
       "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": 9622.474775724513,
-      "required_pages_per_second_upper_estimate": 0.004028729248472873
+      "required_token_positions_per_second": 9646.36658742462,
+      "required_pages_per_second_upper_estimate": 0.004038732251113946
     },
     {
       "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": 19244.949551449026,
-      "required_pages_per_second_upper_estimate": 0.004008215468117687
+      "required_token_positions_per_second": 19292.73317484924,
+      "required_pages_per_second_upper_estimate": 0.004018167536733061
     },
     {
       "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": 28867.42432717354,
-      "required_pages_per_second_upper_estimate": 0.0017305580104093055
+      "required_token_positions_per_second": 28939.09976227386,
+      "required_pages_per_second_upper_estimate": 0.0017348548433015172
     },
     {
       "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": 28867.42432717354,
-      "required_pages_per_second_upper_estimate": 0.002244315267057194
+      "required_token_positions_per_second": 28939.09976227386,
+      "required_pages_per_second_upper_estimate": 0.002249887716869323
     },
     {
       "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": 28867.42432717354,
-      "required_pages_per_second_upper_estimate": 0.37028032384363385
+      "required_token_positions_per_second": 28939.09976227386,
+      "required_pages_per_second_upper_estimate": 0.37119969936601455
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1408848604609716,
-    "producer_p95_seconds": 0.3572087019929313,
-    "gpu_update_seconds": 0.9577577717584708,
-    "producer_batches_per_second": 7.097994750663954,
-    "gpu_batches_per_second": 1.0441053359076078,
-    "producer_headroom": 6.79815963634923,
+    "producer_batch_seconds": 0.08635606080423486,
+    "producer_p95_seconds": 0.3193163340038154,
+    "gpu_update_seconds": 0.9553856279953467,
+    "producer_batches_per_second": 11.579963128088405,
+    "gpu_batches_per_second": 1.0466977633924284,
+    "producer_headroom": 11.0633303452917,
     "sustainable": true,
-    "buffer_seconds": 60.338739620783656,
+    "buffer_seconds": 60.18929456370684,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300296.43451821635,
-    "upload_bytes_per_second": 5012085.309293936,
+    "generation_bytes_per_second": 317071.240585985,
+    "upload_bytes_per_second": 5092893.276016682,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.4366243043887,
+    "estimated_full_shards_per_hour": 68.0361071890322,
     "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": 1.0920391100007691,
-      "bytes": 1249324,
-      "pages": 36,
-      "shards": 1
-    },
     {
       "seconds": 1.1804772159994172,
       "bytes": 13462589,
@@ -562,6 +556,12 @@
       "bytes": 1104603,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 0.9554288359940983,
+      "bytes": 7500029,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -361,3 +361,4 @@
 {"step": 593200, "loss": 0.6198545098304749, "lr": 0.0003, "grad_norm": 0.16725111607689697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86826753976, "tokens_per_second": 153610.33405605968, "data_wait_s": 1.1969001207035035e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 593300, "loss": 0.5967822074890137, "lr": 0.0003, "grad_norm": 0.1605469513533806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86841502857, "tokens_per_second": 152921.5908893095, "data_wait_s": 1.5127996448427439e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 593400, "loss": 0.5555066466331482, "lr": 0.0003, "grad_norm": 0.15687153383663727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86856286107, "tokens_per_second": 152672.49337253757, "data_wait_s": 1.465600507799536e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 593500, "loss": 0.5857194662094116, "lr": 0.0003, "grad_norm": 0.1917992372402928, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86871096226, "tokens_per_second": 154434.70405764738, "data_wait_s": 1.2700002116616815e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f7e65c19f00089d5174d106f50a07bf871d9c2a308a3a4963db302e0ac0fc564
-size 774480
+oid sha256:2f8cd93cf5d348c7eca08fac7e5336d06ab4d589b9ec11d67a4d071b5dc0799f
+size 776598