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": 75552.18958640698,
-      "required_pages_per_second_upper_estimate": 0.005226875780819823
+      "required_token_positions_per_second": 76469.47066961932,
+      "required_pages_per_second_upper_estimate": 0.0052903354145417995
     },
     {
       "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": 9444.023698300873,
-      "required_pages_per_second_upper_estimate": 0.003954015508837844
+      "required_token_positions_per_second": 9558.683833702415,
+      "required_pages_per_second_upper_estimate": 0.004002021313154569
     },
     {
       "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": 18888.047396601745,
-      "required_pages_per_second_upper_estimate": 0.003933882161405761
+      "required_token_positions_per_second": 19117.36766740483,
+      "required_pages_per_second_upper_estimate": 0.003981643526231819
     },
     {
       "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": 28332.071094902618,
-      "required_pages_per_second_upper_estimate": 0.0016984644015717182
+      "required_token_positions_per_second": 28676.051501107242,
+      "required_pages_per_second_upper_estimate": 0.0017190855016960167
     },
     {
       "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": 28332.071094902618,
-      "required_pages_per_second_upper_estimate": 0.0022026939080181385
+      "required_token_positions_per_second": 28676.051501107242,
+      "required_pages_per_second_upper_estimate": 0.002229436871590641
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28332.071094902618,
-      "required_pages_per_second_upper_estimate": 0.3634133873975785
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28676.051501107242,
+      "required_pages_per_second_upper_estimate": 0.3678255987109868
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.45325259092214765,
-    "producer_p95_seconds": 0.26878805000160355,
-    "gpu_update_seconds": 0.9758552386582958,
-    "producer_batches_per_second": 2.2062753087974376,
-    "gpu_batches_per_second": 1.0247421547635496,
-    "producer_headroom": 2.1530053180124287,
+    "producer_batch_seconds": 0.10305808135925076,
+    "producer_p95_seconds": 0.26248719899740536,
+    "gpu_update_seconds": 0.96414947500469,
+    "producer_batches_per_second": 9.703266224354538,
+    "gpu_batches_per_second": 1.0371835757055572,
+    "producer_headroom": 9.355399036042169,
     "sustainable": true,
-    "buffer_seconds": 60.503024796814344,
+    "buffer_seconds": 60.74141692529547,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 312701.1176121372,
-    "upload_bytes_per_second": 1867524.5728969113,
+    "generation_bytes_per_second": 300545.4582743273,
+    "upload_bytes_per_second": 1891746.2260284864,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.0983805301007,
+    "estimated_full_shards_per_hour": 64.49005900547374,
     "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": 1.032158955000341,
-      "bytes": 6117925,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 122.56198037699869,
       "bytes": 6934679,
@@ -562,6 +556,12 @@
       "bytes": 7709457,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 2.9601359040025272,
+      "bytes": 15395500,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -214,3 +214,4 @@
 {"step": 666400, "loss": 0.615656316280365, "lr": 0.0001, "grad_norm": 0.19269532786653254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97544316740, "tokens_per_second": 153093.79737789245, "data_wait_s": 1.0648000170476735e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 666500, "loss": 0.5665801763534546, "lr": 0.0001, "grad_norm": 0.2004456924753406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97558969927, "tokens_per_second": 151801.83431139376, "data_wait_s": 1.2847998732468113e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 666600, "loss": 0.5367074012756348, "lr": 0.0001, "grad_norm": 0.19993975317060272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97573544513, "tokens_per_second": 151572.8148915651, "data_wait_s": 1.4500998076982796e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 666700, "loss": 0.5609776973724365, "lr": 0.0001, "grad_norm": 0.1883014932611547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97588053446, "tokens_per_second": 150949.39067962268, "data_wait_s": 8.032002369873226e-06, "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 sha256:40b6a63f9ca527f2591648dfe3db1c156bbf2e9a40b09d0c6c2277f6d2bb5e9c
-size 460785
+oid sha256:32d7222b65e1913496a802861bdbd85e58a8cdcb589132993494d0ac16d931c4
+size 462903