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": 76513.06297765266,
-      "required_pages_per_second_upper_estimate": 0.005293351231559637
+      "required_token_positions_per_second": 76978.31107587164,
+      "required_pages_per_second_upper_estimate": 0.005325538174518743
     },
     {
       "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": 9564.132872206583,
-      "required_pages_per_second_upper_estimate": 0.004004302711787401
+      "required_token_positions_per_second": 9622.288884483954,
+      "required_pages_per_second_upper_estimate": 0.004028651419692302
     },
     {
       "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": 19128.265744413166,
-      "required_pages_per_second_upper_estimate": 0.003983913308271038
+      "required_token_positions_per_second": 19244.57776896791,
+      "required_pages_per_second_upper_estimate": 0.004008138035631433
     },
     {
       "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": 28692.39861661975,
-      "required_pages_per_second_upper_estimate": 0.0017200654863103915
+      "required_token_positions_per_second": 28866.866653451867,
+      "required_pages_per_second_upper_estimate": 0.0017305245787212096
     },
     {
       "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": 28692.39861661975,
-      "required_pages_per_second_upper_estimate": 0.002230707787918376
+      "required_token_positions_per_second": 28866.866653451867,
+      "required_pages_per_second_upper_estimate": 0.002244271910378281
     },
     {
       "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": 28692.39861661975,
-      "required_pages_per_second_upper_estimate": 0.3680352819566161
+      "required_token_positions_per_second": 28866.866653451867,
+      "required_pages_per_second_upper_estimate": 0.3702731706039156
     }
   ],
   "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": 0.0912899189609746,
-    "producer_p95_seconds": 0.1748207470000125,
-    "gpu_update_seconds": 0.963600163563363,
-    "producer_batches_per_second": 10.954112035387922,
-    "gpu_batches_per_second": 1.0377748342238047,
-    "producer_headroom": 10.555384148991205,
+    "producer_batch_seconds": 0.0524874614921913,
+    "producer_p95_seconds": 0.1759071619999304,
+    "gpu_update_seconds": 0.9577762745058402,
+    "producer_batches_per_second": 19.05216925281808,
+    "gpu_batches_per_second": 1.0440851654170957,
+    "producer_headroom": 18.24771568821882,
     "sustainable": true,
-    "buffer_seconds": 60.70681030449187,
+    "buffer_seconds": 60.339905293867936,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 405979.9121988538,
+    "generation_bytes_per_second": 419949.5209222091,
     "upload_bytes_per_second": 4154639.3647547807,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.11383842920505,
+    "estimated_full_shards_per_hour": 90.11139126538949,
     "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-v1/1790868774355361490-de221e1d/train.jsonl

@@ -59,3 +59,4 @@
 {"step": 446100, "loss": 0.6308509111404419, "lr": 0.0003, "grad_norm": 0.16924410776937326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65295586352, "tokens_per_second": 152139.65855652804, "data_wait_s": 1.1365000318619423e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 446200, "loss": 0.6126097440719604, "lr": 0.0003, "grad_norm": 0.1817995502914307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65310289216, "tokens_per_second": 152658.27359323791, "data_wait_s": 2.5025000468303915e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 446300, "loss": 0.5392016172409058, "lr": 0.0003, "grad_norm": 0.14650099986098924, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65324812221, "tokens_per_second": 151105.54494683084, "data_wait_s": 1.3013999705435708e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 446400, "loss": 0.6222230195999146, "lr": 0.0003, "grad_norm": 0.1682008257322866, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65339327213, "tokens_per_second": 151532.58762522263, "data_wait_s": 1.3633999515150208e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ea5ff0e6b203760a85c9e859a05f76bc99ae71e8c30bc3994be54729cf2aacbf
-size 130146
+oid sha256:09983ade8e989427e921570800a7b79fb118f5321d5b56977fbee46a7f32b361
+size 132264