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": 76907.01604209979,
-      "required_pages_per_second_upper_estimate": 0.005320605818655136
+      "required_token_positions_per_second": 77049.7492428727,
+      "required_pages_per_second_upper_estimate": 0.005330480432671261
     },
     {
       "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": 9613.377005262473,
-      "required_pages_per_second_upper_estimate": 0.004024920201963468
+      "required_token_positions_per_second": 9631.218655359087,
+      "required_pages_per_second_upper_estimate": 0.004032390128282899
     },
     {
       "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": 19226.754010524946,
-      "required_pages_per_second_upper_estimate": 0.004004425816791883
+      "required_token_positions_per_second": 19262.437310718175,
+      "required_pages_per_second_upper_estimate": 0.004011857707190225
     },
     {
       "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": 28840.13101578742,
-      "required_pages_per_second_upper_estimate": 0.001728921818066181
+      "required_token_positions_per_second": 28893.65596607726,
+      "required_pages_per_second_upper_estimate": 0.0017321305571081898
     },
     {
       "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": 28840.13101578742,
-      "required_pages_per_second_upper_estimate": 0.0022421933321475126
+      "required_token_positions_per_second": 28893.65596607726,
+      "required_pages_per_second_upper_estimate": 0.002246354661601172
     },
     {
       "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": 28840.13101578742,
-      "required_pages_per_second_upper_estimate": 0.36993023455044727
+      "required_token_positions_per_second": 28893.65596607726,
+      "required_pages_per_second_upper_estimate": 0.3706167951421513
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08053110156248522,
-    "producer_p95_seconds": 0.18870679199972074,
-    "gpu_update_seconds": 0.9586641608828048,
-    "producer_batches_per_second": 12.417562663340522,
-    "gpu_batches_per_second": 1.0431181646335148,
-    "producer_headroom": 11.904272290860987,
+    "producer_batch_seconds": 0.10277374150770413,
+    "producer_p95_seconds": 0.19498906399894622,
+    "gpu_update_seconds": 0.9568882536865626,
+    "producer_batches_per_second": 9.730111848900995,
+    "gpu_batches_per_second": 1.0450541075693454,
+    "producer_headroom": 9.310629735269805,
     "sustainable": true,
-    "buffer_seconds": 60.395842135616704,
+    "buffer_seconds": 60.28395998225344,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 428772.4892177954,
-    "upload_bytes_per_second": 5694872.885711405,
+    "generation_bytes_per_second": 530859.3878827139,
+    "upload_bytes_per_second": 5744454.111202391,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.00459487343213,
+    "estimated_full_shards_per_hour": 113.9100668655497,
     "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": 1.1198342459999822,
-      "bytes": 518226,
-      "pages": 18,
-      "shards": 1
-    },
     {
       "seconds": 1.3007597939999869,
       "bytes": 12269409,
@@ -562,6 +556,12 @@
       "bytes": 6132473,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.8308523640007479,
+      "bytes": 2627309,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -91,3 +91,4 @@
 {"step": 566200, "loss": 0.5574049353599548, "lr": 0.0003, "grad_norm": 0.164634143063845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82874788475, "tokens_per_second": 153979.27378567585, "data_wait_s": 1.4764000297873281e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 566300, "loss": 0.5719775557518005, "lr": 0.0003, "grad_norm": 0.17521623141029388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82889345800, "tokens_per_second": 151561.8097086228, "data_wait_s": 1.414399957866408e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 566400, "loss": 0.5922015309333801, "lr": 0.0003, "grad_norm": 0.16497290158156128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82903885428, "tokens_per_second": 151594.65350630975, "data_wait_s": 1.4590001228498295e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 566500, "loss": 0.5688738822937012, "lr": 0.0003, "grad_norm": 0.16088098542192927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82918429377, "tokens_per_second": 151430.97849176353, "data_wait_s": 1.3057999240118079e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0f1d727cd850ab192c524774727955d87966bfb3f07145bd933baeb7e5125c64
-size 198705
+oid sha256:ff6be95caf9407c5aa2103c087af83dfe3fa0da58dcdc76ad9a6667893e04620
+size 200823