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": 77489.30938445704,
-      "required_pages_per_second_upper_estimate": 0.005360890223186111
+      "required_token_positions_per_second": 76632.52307124846,
+      "required_pages_per_second_upper_estimate": 0.0053016157580202
     },
     {
       "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": 9686.16367305713,
-      "required_pages_per_second_upper_estimate": 0.004055394459810627
+      "required_token_positions_per_second": 9579.065383906058,
+      "required_pages_per_second_upper_estimate": 0.004010554642609668
     },
     {
       "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": 19372.32734611426,
-      "required_pages_per_second_upper_estimate": 0.004034744903568114
+      "required_token_positions_per_second": 19158.130767812116,
+      "required_pages_per_second_upper_estimate": 0.003990133405051409
     },
     {
       "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": 29058.491019171386,
-      "required_pages_per_second_upper_estimate": 0.00174201216685263
+      "required_token_positions_per_second": 28737.196151718177,
+      "required_pages_per_second_upper_estimate": 0.0017227510301376667
     },
     {
       "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": 29058.491019171386,
-      "required_pages_per_second_upper_estimate": 0.0022591698619464657
+      "required_token_positions_per_second": 28737.196151718177,
+      "required_pages_per_second_upper_estimate": 0.002234190599228746
     },
     {
       "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": 29058.491019171386,
-      "required_pages_per_second_upper_estimate": 0.372731122217152
+      "required_token_positions_per_second": 28737.196151718177,
+      "required_pages_per_second_upper_estimate": 0.3686098966370131
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.2121605929770567,
-    "producer_p95_seconds": 0.34904425499553327,
-    "gpu_update_seconds": 0.951460279949127,
-    "producer_batches_per_second": 4.7134106573134495,
-    "gpu_batches_per_second": 1.051016023552206,
-    "producer_headroom": 4.484623023522653,
+    "producer_batch_seconds": 0.1899474762265072,
+    "producer_p95_seconds": 0.33408611800405197,
+    "gpu_update_seconds": 0.9620980367754822,
+    "producer_batches_per_second": 5.264613249230683,
+    "gpu_batches_per_second": 1.0393951154411956,
+    "producer_headroom": 5.065074071467032,
     "sustainable": true,
-    "buffer_seconds": 60.893457916744126,
+    "buffer_seconds": 60.61217631685538,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 396345.9077563518,
-    "upload_bytes_per_second": 3723199.2133825305,
+    "generation_bytes_per_second": 381656.116714488,
+    "upload_bytes_per_second": 3787701.316098853,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.04660534398953,
+    "estimated_full_shards_per_hour": 81.8945181472395,
     "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": 0.8562311600035173,
-      "bytes": 2647079,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.054913633997785,
       "bytes": 5917684,
@@ -562,6 +556,12 @@
       "bytes": 955502,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.3968777099944418,
+      "bytes": 12101140,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -395,3 +395,4 @@
 {"step": 524500, "loss": 0.6034004092216492, "lr": 0.0003, "grad_norm": 0.1641275472761367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76769183056, "tokens_per_second": 153989.75978307574, "data_wait_s": 9.735995263326913e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 524600, "loss": 0.5592507123947144, "lr": 0.0003, "grad_norm": 0.15388680595723428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76783892664, "tokens_per_second": 154076.70086355964, "data_wait_s": 1.5998994058463722e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 524700, "loss": 0.5254177451133728, "lr": 0.0003, "grad_norm": 0.17957341772498586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76798479936, "tokens_per_second": 151887.0193837863, "data_wait_s": 1.7935999494511634e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 524800, "loss": 0.5686387419700623, "lr": 0.0003, "grad_norm": 0.15876198911507197, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76812983798, "tokens_per_second": 151723.5168645376, "data_wait_s": 1.5948993677739054e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3510479f26ccc56eb46a420ab008148b4454c78fcb3dc873bc008ca5c7af7ea2
-size 846492
+oid sha256:d75575e09550f81fe5dfe049d7bc402d3b26d8811ed974a89c4f8616c6697c82
+size 848610