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": 77096.43351078725,
-      "required_pages_per_second_upper_estimate": 0.005333710158648015
+      "required_token_positions_per_second": 74769.90400347674,
+      "required_pages_per_second_upper_estimate": 0.005172755449040077
     },
     {
       "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": 9637.054188848406,
-      "required_pages_per_second_upper_estimate": 0.004034833344295081
+      "required_token_positions_per_second": 9346.238000434592,
+      "required_pages_per_second_upper_estimate": 0.003913074679138808
     },
     {
       "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": 19274.10837769681,
-      "required_pages_per_second_upper_estimate": 0.004014288482655141
+      "required_token_positions_per_second": 18692.476000869185,
+      "required_pages_per_second_upper_estimate": 0.003893149797239205
     },
     {
       "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": 28911.16256654522,
-      "required_pages_per_second_upper_estimate": 0.001733180051075213
+      "required_token_positions_per_second": 28038.714001303775,
+      "required_pages_per_second_upper_estimate": 0.0016808780917408135
     },
     {
       "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": 28911.16256654522,
-      "required_pages_per_second_upper_estimate": 0.0022477157227841584
+      "required_token_positions_per_second": 28038.714001303775,
+      "required_pages_per_second_upper_estimate": 0.002179886684332316
     },
     {
       "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": 28911.16256654522,
-      "required_pages_per_second_upper_estimate": 0.37084135101583127
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28038.714001303775,
+      "required_pages_per_second_upper_estimate": 0.359650517583199
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09902214268061016,
-    "producer_p95_seconds": 0.361681738999323,
-    "gpu_update_seconds": 0.9563088283413792,
-    "producer_batches_per_second": 10.098751379531734,
-    "gpu_batches_per_second": 1.0456873034774747,
-    "producer_headroom": 9.65752509947088,
+    "producer_batch_seconds": 0.15291299249230406,
+    "producer_p95_seconds": 0.37002678999851923,
+    "gpu_update_seconds": 0.9860651953835826,
+    "producer_batches_per_second": 6.539666667306435,
+    "gpu_batches_per_second": 1.0141317274777117,
+    "producer_headroom": 6.4485376900410225,
     "sustainable": true,
-    "buffer_seconds": 60.24745618550689,
+    "buffer_seconds": 60.149976918398536,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 551813.739166606,
-    "upload_bytes_per_second": 5090967.983494044,
+    "generation_bytes_per_second": 546285.5550543927,
+    "upload_bytes_per_second": 5100593.337075735,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 118.40638285874019,
+    "estimated_full_shards_per_hour": 117.22016323779903,
     "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": 0.962299234000966,
-      "bytes": 720156,
-      "pages": 23,
-      "shards": 1
-    },
     {
       "seconds": 1.5615880700061098,
       "bytes": 13769456,
@@ -562,6 +556,12 @@
       "bytes": 5756789,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.215349038997374,
+      "bytes": 2821977,
+      "pages": 39,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -435,3 +435,4 @@
 {"step": 968500, "loss": 0.5100990533828735, "lr": 3e-05, "grad_norm": 0.338577821665479, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141762164178, "tokens_per_second": 151992.21728515413, "data_wait_s": 1.980800152523443e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 968600, "loss": 0.5253740549087524, "lr": 3e-05, "grad_norm": 0.35511977396952704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141776659980, "tokens_per_second": 150906.8296378775, "data_wait_s": 1.0513998859096318e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 968700, "loss": 0.5292211771011353, "lr": 3e-05, "grad_norm": 0.35714355296152034, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141791288879, "tokens_per_second": 150314.5316953439, "data_wait_s": 8.096001693047583e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 968800, "loss": 0.5146040320396423, "lr": 3e-05, "grad_norm": 0.35547420129965146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141805994106, "tokens_per_second": 153702.53724853892, "data_wait_s": 1.4299999747890979e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1fcf08d333a8c1f81e4c28feb81284b6c3c068deeeaf4a6d23954982ff7e619f
-size 931995
+oid sha256:20d1eeccc614f9b31360e48c1d9f26f3c891b03b8eed34ad9b192203b01acb2a
+size 934113