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": 77374.63002398406,
-      "required_pages_per_second_upper_estimate": 0.005352956438935815
+      "required_token_positions_per_second": 75544.53809720067,
+      "required_pages_per_second_upper_estimate": 0.005226346433042629
     },
     {
       "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": 9671.828752998008,
-      "required_pages_per_second_upper_estimate": 0.004049392728129044
+      "required_token_positions_per_second": 9443.067262150083,
+      "required_pages_per_second_upper_estimate": 0.003953615068994167
     },
     {
       "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": 19343.657505996016,
-      "required_pages_per_second_upper_estimate": 0.004028773731946012
+      "required_token_positions_per_second": 18886.134524300167,
+      "required_pages_per_second_upper_estimate": 0.0039334837605511765
     },
     {
       "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": 29015.486258994024,
-      "required_pages_per_second_upper_estimate": 0.001739434097144464
+      "required_token_positions_per_second": 28329.201786450252,
+      "required_pages_per_second_upper_estimate": 0.0016982923909111792
     },
     {
       "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": 29015.486258994024,
-      "required_pages_per_second_upper_estimate": 0.0022558264310006252
+      "required_token_positions_per_second": 28329.201786450252,
+      "required_pages_per_second_upper_estimate": 0.0022024708319067212
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 29015.486258994024,
-      "required_pages_per_second_upper_estimate": 0.37217950332851074
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28329.201786450252,
+      "required_pages_per_second_upper_estimate": 0.3633765829895749
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.41697643942177365,
-    "producer_p95_seconds": 0.3573520589998225,
-    "gpu_update_seconds": 0.9528704690044566,
-    "producer_batches_per_second": 2.39821703448452,
-    "gpu_batches_per_second": 1.0494605851777352,
-    "producer_headroom": 2.2851901904237413,
+    "producer_batch_seconds": 0.408120429586063,
+    "producer_p95_seconds": 0.3470004289993085,
+    "gpu_update_seconds": 0.9759540776480308,
+    "producer_batches_per_second": 2.450257148396742,
+    "gpu_batches_per_second": 1.0246383747992711,
+    "producer_headroom": 2.3913384552640364,
     "sustainable": true,
-    "buffer_seconds": 60.03083954728076,
+    "buffer_seconds": 60.50915281417791,
     "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": 337542.9392610297,
-    "upload_bytes_per_second": 4951452.135807956,
+    "generation_bytes_per_second": 310359.3584259294,
+    "upload_bytes_per_second": 4955804.746962364,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.42885716794174,
+    "estimated_full_shards_per_hour": 66.59589352210438,
     "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": 1.5840121689980151,
-      "bytes": 6547762,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.2987593140060198,
       "bytes": 2393188,
@@ -562,6 +556,12 @@
       "bytes": 13104812,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 1.2835482930022408,
+      "bytes": 5442288,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -417,3 +417,4 @@
 {"step": 686700, "loss": 0.5152714252471924, "lr": 0.0001, "grad_norm": 0.19509901156734033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100514906071, "tokens_per_second": 151746.89700887015, "data_wait_s": 1.611899642739445e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 686800, "loss": 0.5629808902740479, "lr": 0.0001, "grad_norm": 0.20279141272276377, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100529481032, "tokens_per_second": 145259.7477520443, "data_wait_s": 9.51300171436742e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 686900, "loss": 0.5643567442893982, "lr": 0.0001, "grad_norm": 0.20095767760768526, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100544126019, "tokens_per_second": 140151.1358112805, "data_wait_s": 1.2346994481049478e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 687000, "loss": 0.5700095295906067, "lr": 0.0001, "grad_norm": 0.2111146027716485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100558815263, "tokens_per_second": 152244.84273307477, "data_wait_s": 1.2853997759521008e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:824f974a7e4229b3b4df5ee8b93008f0b3e7e8f4958cb6ee144c453a5cdb911e
-size 893871
+oid sha256:b7b69195abe9f37cc8ece3faf7d71ce0831bde08e6836b13df6246a972c7ea93
+size 895989