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": 76641.08489937382,
-      "required_pages_per_second_upper_estimate": 0.005302208085156091
+      "required_token_positions_per_second": 77306.57796973751,
+      "required_pages_per_second_upper_estimate": 0.0053482484399205174
     },
     {
       "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": 9580.135612421727,
-      "required_pages_per_second_upper_estimate": 0.004011002724940266
+      "required_token_positions_per_second": 9663.322246217189,
+      "required_pages_per_second_upper_estimate": 0.004045831231375976
     },
     {
       "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": 19160.271224843455,
-      "required_pages_per_second_upper_estimate": 0.003990579205803389
+      "required_token_positions_per_second": 19326.644492434378,
+      "required_pages_per_second_upper_estimate": 0.004025230369884462
     },
     {
       "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": 28740.40683726518,
-      "required_pages_per_second_upper_estimate": 0.0017229435058337763
+      "required_token_positions_per_second": 28989.966738651565,
+      "required_pages_per_second_upper_estimate": 0.0017379042408659847
     },
     {
       "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": 28740.40683726518,
-      "required_pages_per_second_upper_estimate": 0.0022344402159077126
+      "required_token_positions_per_second": 28989.966738651565,
+      "required_pages_per_second_upper_estimate": 0.002253842400542506
     },
     {
       "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": 28740.40683726518,
-      "required_pages_per_second_upper_estimate": 0.3686510798638445
+      "required_token_positions_per_second": 28989.966738651565,
+      "required_pages_per_second_upper_estimate": 0.3718521663222838
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0584353383124494,
-    "producer_p95_seconds": 0.2013043370006926,
-    "gpu_update_seconds": 0.9619905576336953,
-    "producer_batches_per_second": 17.11293249733705,
-    "gpu_batches_per_second": 1.0395112426672881,
-    "producer_headroom": 16.462479475861052,
+    "producer_batch_seconds": 0.4319589062500029,
+    "producer_p95_seconds": 0.2138562140007707,
+    "gpu_update_seconds": 0.9537092694603766,
+    "producer_batches_per_second": 2.315035031182421,
+    "gpu_batches_per_second": 1.0485375701190525,
+    "producer_headroom": 2.207870368364167,
     "sustainable": true,
-    "buffer_seconds": 60.605405130922804,
+    "buffer_seconds": 60.08368397600373,
     "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": 356515.75240393716,
+    "generation_bytes_per_second": 323973.66989901883,
     "upload_bytes_per_second": 5853710.994822643,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.49998120392405,
+    "estimated_full_shards_per_hour": 69.51720783927844,
     "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": [

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

@@ -132,3 +132,4 @@
 {"step": 498200, "loss": 0.5864498615264893, "lr": 0.0003, "grad_norm": 0.1551933277041455, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72919966630, "tokens_per_second": 152959.19352431633, "data_wait_s": 1.6435998986708e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 498300, "loss": 0.5782458782196045, "lr": 0.0003, "grad_norm": 0.1476533573606139, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72934606497, "tokens_per_second": 152606.39386160494, "data_wait_s": 9.33699993765913e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 498400, "loss": 0.5458378195762634, "lr": 0.0003, "grad_norm": 0.16699920113869673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72949286780, "tokens_per_second": 141572.82226803506, "data_wait_s": 1.2077000064891763e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 498500, "loss": 0.5711762309074402, "lr": 0.0003, "grad_norm": 0.1529487524209411, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72963879918, "tokens_per_second": 151916.1781062014, "data_wait_s": 1.2923999747727066e-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 sha256:1568c25012353b32b1342f7be1f05811de6d9262222e9e9c7dd4f8c345569604
-size 285543
+oid sha256:1ea713820685ef96a1f3a60ac36c693e92100f6638d3d2cf9d20bbc94912032d
+size 287661